JP7049629B2 - Voice concealment device and voice concealment program - Google Patents

Description

The present invention relates to a voice concealment device and a voice concealment program.

In recent years, information terminal devices (for example, smartphones, smart speakers, etc.) that automatically analyze voices emitted by users using voice recognition technology and provide services for searching and acquiring information requested by users by connecting to the Internet. Smart home appliances and interactive robots, including) are becoming more widespread. There is a risk that personal information (home address, telephone number, etc.) issued during conversations in homes, public facilities, workplaces, etc. where such information terminal devices exist may be unintentionally leaked onto the network. ing. For example, when a user of an information terminal device (for example, a child or an elderly person) speaks without being aware of personal information, or a voice containing personal information emitted by a person who is unaware of the existence of the information terminal device is transmitted to the information terminal device. Personal information may be unintentionally leaked to the outside, such as when it is entered incorrectly. Therefore, there is a demand for a technique for preventing the leakage of personal information in an information terminal device having a voice recognition function. For example, Patent Document 1 discloses a technique for deleting personal information contained in an audio signal.

Japanese Unexamined Patent Publication No. 2010-091761

However, the conventional technology cannot prevent the outflow of the information at the stage when the voice input to the information terminal device may match the information that the user wants to keep confidential, such as personal information. There is a problem. As a result, when it is found that the input voice contains personal information, there is a possibility that an undesired situation may occur in which the voice has already leaked to the network.
The present invention has been made in view of such a point, and when there is a possibility that the voice input to the information terminal device matches the information desired to be concealed by the user, the leakage of the information occurs. It is an object of the present invention to provide a voice concealment device that can be prevented.

In order to achieve the above object, the voice concealment device according to one aspect of the present invention has a storage unit capable of storing phonetic sequence information in which a plurality of keywords are converted into voice information, and digitally inputs voice from a voice source. The voice recognition unit that outputs the voice data converted into a signal as output sound element information which is the information recognized at the sound element level by the sequential voice recognition process, and the output sound element information output by the voice recognition unit and the storage unit are stored. A determination unit capable of sequentially determining whether or not the input voice and the keyword may match based on the degree of matching with the stored phonetic information, which is the phonetic sequence information, and the determination unit match the keyword. It is provided with a voice data concealment unit capable of concealing the contents of voice data corresponding to the input voice determined to be likely to occur, and the determination unit includes output phonetic information output by the voice recognition unit and the storage unit. Based on the degree of prefix matching with the stored phonetic information, which is the phonetic sequence information stored in, it is possible to sequentially determine whether or not the input voice and the keyword may match, and the voice data concealment unit. Is characterized in that the content of the voice data corresponding to the output phonetic information determined by the determination unit to have a degree of prefix matching with the stored phonetic information is equal to or higher than a predetermined degree can be concealed .
Further, in order to achieve the above object, the voice concealment device according to another aspect of the present invention has a storage unit capable of storing phonetic sequence information in which a plurality of keywords are converted into voice information, and a voice source. A voice recognition unit that outputs voice data obtained by converting input voice into a digital signal as output voice element information, which is information recognized at the sound element level by sequential voice recognition processing, output sound element information output by the voice recognition unit, and the storage unit. A determination unit capable of sequentially determining whether or not the input voice and the keyword may match based on the degree of matching with the stored phonetic information stored in the determination unit, and the determination unit . The voice recognition unit includes a voice data concealment unit capable of concealing the content of the voice data corresponding to the input voice determined to be likely to match the keyword, and the voice recognition unit converts the input voice into voice data. The voice data concealment unit discriminates between a voice section in which voice exists and a silent section in which voice does not exist, and the voice data concealment unit is based on the determination by the determination unit that the input voice may match the keyword. , The content of the voice data corresponding to the input voice input from the voice source during a predetermined fixed period after the determination, or the voice data corresponding to the input voice input from the voice source after the judgment. It is characterized in that the contents of the audio data of a predetermined number of audio sections are concealed.

Further, the voice concealment program according to one aspect of the present invention is a storage unit capable of storing a predetermined information terminal device, a storage unit capable of storing phonetic sequence information in which a plurality of keywords are converted into voice information, and digitally input voice from a voice source. The voice recognition unit that outputs the voice data converted into a signal as output voice element information which is the information recognized at the sound element level by the sequential voice recognition process, the output sound element information output by the voice recognition unit, and the storage unit are stored. A determination unit capable of sequentially determining whether or not there is a possibility that the input voice and the keyword match based on the degree of matching with the stored phonetic information which is the sound element sequence information, and the determination unit matches the keyword. The content of the voice data corresponding to the input voice determined to be dangerous is made to function as a voice data concealment unit, and the determination unit stores the output phonetic information output by the voice recognition unit and the storage unit. It is possible to sequentially determine whether or not the input voice and the keyword may match based on the degree of prefix matching with the stored phonetic information which is the phonetic sequence information, and the voice data concealment unit can determine whether or not the input voice and the keyword may match. It is characterized in that the content of the voice data corresponding to the output phonetic information determined by the determination unit to have a degree of prefix matching with the stored phonetic information is equal to or higher than a predetermined degree can be concealed .

According to one aspect of the present invention, it is possible to prevent the leakage of the information at the stage when the voice input to the information terminal device may match the information that the user wants to keep confidential, such as personal information. ..

It is a schematic block diagram of the voice recognition system including the voice concealment device of 1st Embodiment of this invention. It is a flowchart which shows the flow of the voice concealment processing of 1st Embodiment of this invention. It is a figure explaining an example of the confidential information determination processing of 1st Embodiment of this invention in time series. It is a figure explaining another example of the confidential information determination processing of 1st Embodiment of this invention in time series. It is a figure explaining an example of the confidential information determination processing in the modification of 1st Embodiment of this invention in time series. It is a figure explaining another example of the confidential information determination processing in the modification of 1st Embodiment of this invention in time series. It is a schematic block diagram of the voice concealment apparatus of the 2nd Embodiment of this invention. It is a flowchart which shows the flow of the voice concealment processing of the 2nd Embodiment of this invention.

1. 1. First Embodiment The voice concealment device according to the first embodiment of the present invention will be described with reference to FIGS. 1 to 3. First, the configuration of the voice recognition system including the voice concealment device according to the first embodiment will be described with reference to FIG.

(1-1) Configuration of Voice Recognition Service System FIG. 1 is a diagram showing a schematic configuration of a voice recognition service system 9 including a voice concealment device 10 according to the present embodiment. The voice recognition service system 9 includes a voice processing device 1 including a voice concealment device 10, a voice recognition server 100 connected to the voice processing device 1 via a network 3 such as the Internet, and a server connected to the voice recognition server 100. Consists of 500. The server 500 is, for example, a server that constitutes a search system (search engine). The voice recognition service system 9 is a system that provides a service for searching and acquiring information requested by a user by voice.

When the voice recognition server 100 receives voice data (for example, voice data based on the user's speech) transmitted from the voice processing device 1 via the network 3, the voice recognition server 100 performs voice recognition on the received voice data, and the result of this voice recognition. To the server 500. Based on the transmitted voice recognition result, the server 500 performs processing such as information retrieval to acquire the information requested by the user, and returns the acquired information (processing result) to the voice processing device 1 via the network 3. .. As a result, the voice processing device 1 can output the information returned from the server 500 to the user. In this way, the voice recognition service system 9 can provide a service (for example, an information retrieval service) corresponding to the voice input from the user.

[Voice processing device]
As shown in FIG. 1, the voice processing device 1 includes a voice input device 11, a voice concealment device 10, and a dialogue processing device 110. The voice processing device 1 is a device (for example, an information terminal device) having a voice recognition function and a network communication function of, for example, a smartphone, a smart speaker, an interactive robot, a smart home appliance, or the like. The voice processing device 1 analyzes the sound input from the sound collecting device (mic or the like) in the voice input device 11 and converts it into digital voice data, and outputs the digital voice data to the network 3 via the voice concealment device 10. do. As a result, the voice processing device 1 can transmit the input voice data to the voice recognition server 100 via the network 3. The voice processing device 1 does not have to have a network communication function. For example, the voice processing device 1 may be a device that provides services that do not depend on connection to a network (for example, alarm, dialogue, memo function, etc.).

The voice input device 11 can input voice data (voice signal) generated by a predetermined voice source. Here, the voice data input to the voice input device 11 (hereinafter referred to as "input voice data") is assumed to be, for example, the voice spoken by the user of the voice processing device 1 (input voice). The voice generation source is not limited to the user of the voice processing device 1, and may be, for example, a device that emits voice. The voice input device 11 includes, for example, a sound collecting device (not shown) such as a microphone. The audio input device 11 samples input audio data (analog audio data) input as an analog signal from the sound pickup device at a specified sampling frequency, performs analog-to-digital (AD) conversion, and performs analog-to-digital (AD) conversion, and digital signals (digital). Audio data) is generated.

The voice input device 11 outputs digital voice data generated from the input voice data (analog voice data) to the voice concealment device 10. Specifically, the voice input device 11 outputs the generated digital voice data to the voice concealment device 10 in sampling units at the time of AD conversion. The digital audio data in the sampling unit is in a state where the time series is retained. Further, in this example, the voice input device 11 will be described as one element constituting the voice processing device 1, but the present invention is not limited to this. The voice input device 11 may be a component of the voice concealment device 10.

[Voice concealment device]
The voice concealment device 10 according to the present embodiment may refer to information (hereinafter, "confidential information") that the user of the voice processing device 1 desires to conceal among the voice (input voice data) input to the voice processing device 1. ) May be included Conceal input audio data. Here, concealment of input voice data means to stop the output of voice data that may contain confidential information from the voice processing device 1. As a result, the voice concealment device 10 can prevent the concealment information from leaking from the voice processing device 1 via the network 3. In this example, the information assumed as confidential information is, for example, personal information. Here, personal information includes information that can identify an individual (address, telephone number, birthday, various passwords, etc.) and information on financial assets (credit card or cash card identification number, bank account number, etc.). Can be mentioned. The confidential information to be prevented from being leaked by the voice concealment device 10 of the present invention is not limited to the personal information listed here. The voice concealment device 10 can prevent the leakage of arbitrary information registered in advance by the user of the voice processing device 1 as concealment information.

Hereinafter, the voice concealment device 10 will be described in detail with reference to FIG.
The voice concealment device 10 includes a voice recognition unit 12, a determination unit 13, a storage unit 14, and a voice data concealment unit 15. The digital voice data generated from the input voice data in the voice input device 11 of the voice processing device 1 is input to the voice recognition unit 12 of the voice concealment device 10 and also input to the voice data concealment unit 15.

The voice recognition unit 12 performs voice recognition in real time (sequentially in chronological order) each time the voice input device 11 samples and outputs digital voice data. Hereinafter, this voice recognition process performed in real time may be referred to as "sequential voice recognition process". The voice recognition unit 12 has an engine unit 120 that executes sequential voice recognition processing, and a model 121 that the engine unit 120 uses for voice recognition. The engine unit 120 recognizes digital voice data at the phoneme level. Here, a phoneme is the minimum unit that constitutes the pronunciation of a language, and is, for example, the minimum value of the unit of sound that constitutes a beat (mora) in the pronunciation of Japanese. The voice recognition unit 12 can recognize digital voice data based on voices of various languages, not limited to Japanese, at the phoneme level.

The determination unit 13 uses input voice data (mainly a user) input to the voice input device 11 using phoneme information (hereinafter, may be referred to as “output phoneme”) which is digital voice data recognized at the phoneme level. (Spoken voice) determines whether or not there is a possibility that confidential information is included (confidential information determination processing). In the voice concealment device 10 according to the present embodiment, the concealed information used for the determination process by the determination unit 13 is stored in the concealment information list 141 stored in the storage unit 14 which is a predetermined storage area such as a memory.

The confidential information list 141 stores a plurality of confidential information (for example, personal information) in a table format, for example. The confidential information stored in the confidential information list 141 is information registered in advance by the user of the voice processing device 1 using a predetermined information terminal device (smartphone, personal computer, etc.) connected to the voice processing device 1. The information terminal device transmits information (hereinafter, may be referred to as a "keyword") for registering in the confidential information list 141, for example, text-input by the user, to the voice processing device 1.

The keyword is confidential information that the user wants to keep confidential, and includes, for example, personal information. The voice concealment device 10 converts the received keyword into phoneme sequence information which is sound information and stores it in the personal information list 141. The user of the voice processing device 1 can arbitrarily register the secret information that is considered to be undesired to be output from the voice processing device 1 to the voice recognition server 100 in the secret information list. As a result, the voice concealment device 10 according to the present embodiment can prevent the information from leaking from the voice processing device 1 via the network. The keyword is confidential information that the user intentionally registers in the confidential information list 141, and may not correspond to personal information as described above. For example, as confidential information, a guardian registers in the confidential information list 141 terms related to information that is considered to be unfavorable or dangerous for children to acquire and use by information retrieval using the voice processing device 1. be able to. As a result, the voice concealment device 10 can prevent the child from inadvertently acquiring dangerous information by using the voice processing device 1.

FIG. 1 shows a telephone number "09012345678" which is a kind of personal information as an example of confidential personal information stored in the confidential information list 141 on the right side of the voice processing device 1. In addition, in FIG. 1, in order to facilitate understanding, the confidential information (keyword) stored as the phoneme sequence information is shown as a character string instead of the phoneme notation. In addition, the telephone number may be input by separating with "-(hyphen)", or may be input by omitting the hyphen as shown in FIG. In addition, the address may be entered separated by "-(hyphen)", and may be entered as "1-chome ...". Furthermore, when speaking a telephone number or address, various utterance patterns are assumed. For example, a pattern in which a telephone number is spoken by inserting a "no" in the hyphenated part such as "090 ...", or a pattern in which an address in an address is spoken as "1 of 2 ...", "1" There are patterns of utterances such as "chome ...". For example, the number "0" is assumed to have multiple utterance patterns such as "zero" and "ray". For this reason, the voice concealment device 10 automatically expands, for example, a part corresponding to a telephone number, an address delimiter (hyphen or space), a number, or the like in a keyword into a plurality of assumed speech patterns in advance, and is a phoneme. It is converted into series information and registered in the confidential information list 141.

The determination unit 13 sequentially executes the confidential information determination process each time the output phoneme information is output from the voice recognition unit 12, and the output phoneme is a phoneme sequence indicating the confidential information (keyword) accumulated in the confidential information list 141. Confidential information judgment result indicating whether or not there is a possibility that confidential information such as personal information may be included in the input voice based on the "degree of matching" with (hereinafter, may be referred to as "confidential information phoneme series"). Is derived. Here, the degree of matching is calculated based on the degree of matching (numerical value) for each phoneme constituting the confidential information phoneme sequence, and for example, when the calculated degree of matching is equal to or higher than a predetermined degree (threshold value). In some cases, it is determined that there is a possibility of matching (confidential information may be included). Also, from the viewpoint of which part of the confidential information phoneme sequence the output phoneme matches, "exact match" (match with all parts), "prefix match" (match with the front part), and "intermediate match". (Matching with the intermediate part), etc., the degree of matching can be calculated by various methods, but the case where the confidential information determination result is derived based on the "prefix matching" will be described below. When a plurality of confidential information is stored in the confidential information list 141, the determination unit 13 derives the confidential information determination result for each of the plurality of confidential information. The determination unit 13 outputs the derived confidential information determination result to the voice data confidential unit 15. In addition, the determination unit 13 also derives a perfect match determination result as to whether or not the combination of output phonemes corresponding to the continuous input voice (combined output phoneme sequence) completely matches the secret information phoneme sequence. The exact match determination result is transmitted to the output control unit 150 as a part of the above-mentioned confidential information determination result. The exact match determination process for deriving the exact match determination result will be described later.

The voice data concealment unit 15 executes a voice concealment process for concealing digital audio data including contents that may correspond to confidential information. As shown in FIG. 1, the voice data concealment unit 15 has an output control unit 150 and a buffer 151. The buffer 151 can temporarily store the digital voice data output from the voice input device 11. Further, the output control unit 150 controls the buffer output of the digital audio data stored in the buffer 151 by determining whether or not the digital audio data can be output based on the confidential information determination result derived by the determination unit 13. In this example, the input / output of digital audio data in the buffer 151 is performed by a first-in first-out (FIFO) method. Details of the buffer output control by the output control unit 150 will be described later.

[Dialogue processing device]
The dialogue processing device 110 sequentially outputs the processing results of the server 500 returned from the network 3 in a format that can be understood by the user, such as text data and voice. As a result, for example, the processing result in the server 500 (for example, the information retrieval result) is provided to the user of the voice processing device 1. Although the dialogue processing device 110 is included in the voice processing device 1 in FIG. 1, it may exist separately from the voice processing device 1.

(1-2) Outline of voice recognition in the voice recognition unit Here, the voice recognition (sequential voice recognition processing) executed by the engine unit 120 of the voice recognition unit 12 will be described. Since the technique of voice recognition by the engine unit 120 is generally used, only the outline thereof will be described in this specification.

The digital voice data generated by the voice input device 11 includes not only the voice spoken by the user but also the environmental sound around the voice processing device 1. Therefore, first, the engine unit 120 of the voice recognition unit 12 determines a section in which voice exists and a section in which voice does not exist from a digital signal (digital voice data) including voice and environmental sound (Voice Activity Detection). : VAD) is performed (voice section detection processing).

Following the voice section detection process, the engine unit 120 extracts the feature amount from the digital voice data in the voice section in time series (acoustic feature extraction process). The engine unit 120 performs time-frequency conversion processing such as Fourier transform on the voice spectrum of the voice section detected in the voice section detection process, and extracts an acoustic feature vector. The engine unit 120 divides the voice section into predetermined time intervals (for example, about 10 milliseconds), and performs acoustic feature extraction processing at each time interval.

Next, the engine unit 120 performs continuous phoneme recognition of digital voice data based on the model 121 and the feature amount (phoneme feature vector) of the voice information extracted in the sound feature extraction process, and recognizes the phoneme level (phoneme level). Recognition process). In this example, the model 121 is a phoneme model to which, for example, a hidden Markov model (HMM) is applied. In the phoneme recognition process, the engine unit 120 collates the digital voice data input from the voice input device 11 with the phoneme using the model 121 by the hypothesis search method, and determines the collated phoneme (output phoneme). Output to.

As described above, in the voice concealment device 10 according to the present embodiment, the input voice based on the user's speech is input to the voice input device 11, and the voice input device 11 converts the input voice data (analog signal) into digital voice data (digital signal). When the data is converted to In this way, the digital voice data (an example of voice data) input from the voice input device 11 is recognized by the voice recognition unit 12 at the phoneme level and output to the determination unit 13.

(1-3) Voice concealment processing Next, an example of the flow of voice concealment processing in the voice data concealment unit 15 will be described with reference to FIG. 1 and FIG. FIG. 2 is a flowchart illustrating a flow of voice concealment processing. The voice concealment process is a process of stopping the output of voice data including confidential information such as personal information from the voice processing device 1 by controlling the buffer output by the output control unit 150. In the present embodiment, the voice concealment processing is executed based on the fact that the concealment information determination processing result is input from the determination unit 13 to the voice data concealment unit 15. That is, the voice concealment processing is repeatedly executed every time the input voice data is input to the voice processing device 1 and output to the voice recognition unit 12 in sampling units, and the concealment information determination processing is executed in the determination unit 13.

(Step S21)
In step S21, the output control unit 150 receives the confidential information determination result from the determination unit 13, and proceeds to the process of step S22.

(Step S22)
In step S22, the output control unit 150 confirms the confidential information determination result based on the confidential information determination signal received from the determination unit 13, and determines whether or not it is possible to output digital voice data from the buffer 151 (buffer output). judge. The confidential information determination signal is a signal indicating whether or not the output phoneme corresponding to the digital audio data currently being processed has a prefix match with the confidential information phoneme sequence, and is generated by the determination unit 13 to conceal the audio data. It is output to unit 15. When the output control unit 150 receives the secret information determination signal indicating that the output phoneme is prefix-matched with the secret information phoneme sequence, the output control unit 150 determines that buffer output is impossible and proceeds to the process of step S23.

The digital voice data to be processed is the digital voice data stored in the buffer 151 at the time of executing the voice concealment processing. As described above, the voice input device 11 simultaneously outputs digital voice data to the voice recognition unit 12 and the buffer 151 in the sampling unit at the time of AD conversion of the input voice data. Further, the output control unit 150 conceals the input voice by the concealment information determination process based on the reception of the concealment information determination signal indicating that the output phoneme coincides with the concealment information phoneme sequence. Recognize that it is determined that information may be included.

On the other hand, when the output control unit 150 receives a secret information determination signal indicating that the output phoneme does not match the secret information phoneme sequence in the front, the output control unit 150 determines that buffer output is possible and proceeds to the process of step S26. .. Based on the fact that the output control unit 150 receives the secret information determination signal indicating that the output phoneme is prefix-matched with the secret information phoneme sequence, the determination unit 13 performs the secret information determination process to display the secret information in the input voice. It is recognized that it is judged that there is no possibility that it is included.

(Step S23)
In step S23, the output control unit 150 stops the output of the digital audio data currently stored in the buffer 151, and proceeds to the process of step S24.

(Step S24)
In step S24, the output control unit 150 has a content that completely matches any of the keywords (confidential information) such as personal information registered in advance by the user in the input voice based on the perfect match determination signal received from the determination unit 13. It is confirmed whether or not it is determined to be included, that is, whether or not it is determined that the combined output phoneme sequence corresponding to the continuous input voice data completely matches any of the secret information phoneme sequences. The perfect match determination signal is a signal indicating whether or not the combined output phoneme sequence completely matches any of the secret information phoneme sequences registered in the secret information list 141, and is a signal generated by the determination unit 13 and is a voice data concealment unit. It is output to 15. When the output control unit 150 receives a perfect match determination signal indicating that the combined output phoneme sequence completely matches any of the secret information phoneme sequences registered in the secret information list 141, the output control unit 150 receives personal information in the input voice. It is determined that the content that completely matches the keyword such as is included, and the process proceeds to step S25. On the other hand, when the output control unit 150 receives an exact match determination signal indicating that the combined output phoneme sequence does not completely match any of the secret information phoneme sequences registered in the secret information list 141, the output control unit 150 receives an exact match determination signal in the input voice. It is determined that the content that completely matches the keyword such as personal information is not included, and the voice concealment process is terminated.

(Step S25)
At present, in step S25, the output control unit 150 receives an exact match determination signal indicating that the input voice contains a content that completely matches a keyword (confidential information) such as personal information. The digital signal data stored in the buffer 151 is deleted, the inside of the buffer 151 is cleared, and the voice concealment process is completed. In the voice concealment device 10 according to the present embodiment, digital signal data having a content that completely matches a keyword including personal information is deleted from the buffer 151, so that information (confidential information) that the user desires to conceal, such as personal information, is deleted. Can be reliably prevented from flowing out.

(Step S26)
In step S26, the output control unit 150 outputs the digital voice data determined by the determination unit 13 that does not match the confidential information registered in the confidential information list 141 from the buffer 151. As a result, the voice concealment device 10 can immediately output digital voice data that does not correspond to the concealment information and transmit it to the voice recognition server 100 via the network 3.

As described above, in the voice concealment device 10 according to the present embodiment, the output control unit (an example of the buffer output control unit) 150 of the voice data concealment unit 15 is determined by the determination unit 13 to be forward-matched with the concealment information phone series. The digital voice data corresponding to the output sound element, that is, the digital voice data corresponding to the input voice data determined to match the keyword such as personal information is not output from the buffer 151. As a result, the voice concealment device 10 according to the present embodiment prevents the concealment information (for example, a keyword including personal information and the like) registered by the user of the voice processing device 1 from being transmitted to the voice recognition server 100 via the network 3. can do.

The voice processing device 1 provided with the voice concealment device 10 always performs voice concealment processing in the voice data concealment unit 15 for the input voice data, and it is determined that there is no possibility of matching (prefix matching) with a keyword such as personal information. Output digital audio data. Therefore, the digital voice data output from the voice processing device 1 does not include confidential information that the user desires to keep secret.

(1-4) Confidential Information Determination Process Next, the details of the confidential information determination process in the determination unit 13 will be described with reference to FIGS. 1 and 2. FIG. 3 is a diagram illustrating an example of the flow of the confidential information determination process in chronological order. In this example, the confidential information phoneme sequence information indicating "090012344567" is registered in the confidential information list 141 as the information corresponding to the telephone number. Further, it is assumed that a total of 10 confidential information phoneme sequences (confidential information 1 to 10) including the above-mentioned telephone number are accumulated in the confidential information list 141. FIG. 3 shows the confidential information when the voice input device 11 inputs a voice (for example, the telephone number is 090012345667) including a content that completely matches the confidential information (for example, the telephone number “090012345667”). An example of the determination process is shown.

More specifically, in the first and second stages from the top of FIG. 3, the correspondence between the input voice in the voice input device 11 and the section containing one or more phonemes in the series of input voices is illustrated. In the third stage of 3, the output phonemes corresponding to each section of the input voice are illustrated. In the input voice, each of the vowel (a, i, u, e, o) and the sound other than the vowel (consonant, etc.) corresponds to one phoneme, and in FIG. 3, each phoneme is expressed in Roman letters. For example, as shown in FIG. 3, the output phoneme "den" corresponding to the input voice "den" is formed of three phonemes "d", "e", and "n". In this example, in order to facilitate the explanation, the input voice is divided into sections including a plurality of phonemes, and the operation of each configuration of the voice concealment device 10 in the sequential voice recognition process and the concealment information determination process executed for this section will be described. do.

Further, in the fourth row of FIG. 3, the progress of collation of output phonemes in the voice recognition unit 12 is shown, and in the fifth row, a time axis showing the passage of time from time t0 to time t5 is shown. ing. Further, in the 6th and 7th columns of FIG. 3, the signal level (high level: High or low level: Low) of the personal information determination signal corresponding to each confidential information (keyword) registered in the confidential information list 141 is displayed. It is illustrated, and in the eighth row, a secret information determination signal (comprehensive determination signal) showing the result of taking the logical sum of the states of the confidential information determination signal is shown. Further, in the 9th and 10th stages of FIG. 3, the signal level (high level: High or low level: Low) of the perfect match signal is shown, and in the 11th stage, the logical sum of the states of the perfect match signal is taken. An exact match signal (comprehensive determination signal) showing the result is shown. In the twelfth stage, the input voice corresponding to the digital voice data stored in the buffer 151 is shown. Further, in the thirteenth stage of FIG. 3, the input voice corresponding to the digital voice data output from the buffer 151 is shown.

Here, the signal levels of the confidential information determination signal and the perfect match signal illustrated in FIG. 3 will be described. The secret information determination signal having a high signal level indicates that the output phoneme is forward-matched with the secret information phoneme sequence as the secret information determination result. Further, the secret information determination signal having a low signal level indicates that the output phoneme does not forwardly match the secret information phoneme sequence as the secret information determination result. In the voice concealment device 10 according to the present embodiment, the concealment information determination signal (comprehensive) shown in the eighth stage of FIG. 3 is output from the determination unit 13 to the voice data concealment unit 15.

Further, the perfect match determination signal having a high signal level indicates that the combined output phoneme sequence completely matches the secret information phoneme sequence. Further, the perfect match determination signal having a low signal level indicates that the combined output phoneme sequence does not completely match the secret information phoneme sequence. In the voice concealment device 10 according to the present embodiment, the exact match signal (comprehensive) shown in the 11th stage of FIG. 3 is output from the determination unit 13 to the voice data concealment unit 15.

The voice input device 11 samples the above-mentioned input voice data (voice signal) input from the microphone, converts it into digital voice data, and sequentially outputs the digital voice data to the voice recognition unit 12 and the buffer 151.

In the voice recognition unit 12, the engine unit 120 sequentially executes voice recognition processing in chronological order for the digital voice data input from the voice input device 11. The engine unit 120 outputs the output phoneme obtained as a result of the phoneme recognition process in the series of sequential speech recognition processes to the determination unit 13. In this example, the digital voice data includes the contents corresponding to the phonemes constituting the above-mentioned input voice (“telephone number is 090012345667”). For this reason, for example, the series of input voices shown in FIG. 3 are divided into sections for ease of understanding: "den", "wa", "van", "go", "wa", "zero", "queue", and "zero". It corresponds to 16 sections including digital voice data corresponding to "Zero", "Ichi", "Knee", "Go", "Yon", "Roku", "Nana", and "Hachi" respectively. Here, as shown in FIG. 3, 16 sections are designated as sections 1 to 16 in chronological order. The digital voice data included in each of the sections 1 to 16 is output from the voice input device 11 to the voice recognition unit 12 in chronological order, and the engine unit 120 of the voice recognition unit 12 outputs the digital voice of each section in the order of input. Sequential voice recognition processing is executed for the data.

As shown in FIG. 3, for example, at time t0, the voice input device 11 outputs digital voice data (corresponding to the input voice “den”) included in the section 1 to the voice recognition unit 12. At this time, the digital voice data (corresponding to the input voice “den”) included in the section 1 is also output to the buffer 151. Similarly, each digital voice data included in the sections 2 to 16 is also output to the voice recognition unit 12 and the buffer 151 at the same time. The voice recognition unit 12 sequentially executes voice recognition processing on the digital voice data included in the section 1, and outputs the phoneme information "den" corresponding to the Japanese utterance "den" to the determination unit 13. Output. For ease of understanding, the output phonemes are described in katakana notation in the specification in the same manner as the input voice except for some parts.

When the output phoneme "den" corresponding to the digital voice data included in the section 1 is input, the determination unit 13 executes the confidential information determination process for the output phoneme "den". Specifically, the determination unit 13 determines whether or not each of the confidential information phoneme sequences stored in the confidential information list 141 has a prefix match with the output phoneme "den", and determines whether or not the determination result is a confidential information determination signal. Is output to the voice data concealment unit 15. As shown in FIG. 3, in this example, the confidential information list 141 includes the confidential information 1 corresponding to the above-mentioned telephone number, and the keywords including all 10 personal information corresponding to the credit card number, the name, etc. ( Confidential information 1 to 10) is converted into a confidential information phone series and stored. The determination unit 13 determines whether or not all the confidential information phoneme sequences in the confidential information list 141 have a prefix match with the output phoneme.

For example, the determination unit 13 includes the secret information phoneme series "Zero Cue Zero Ichiny Sanyon Gourokunana" corresponding to the secret information text "090012344567" registered as the secret information 1 and the output phoneme "Den" output at time t0. Judges that they do not match at the beginning. Therefore, the determination unit 13 sets the confidential information determination signal corresponding to the confidential information 1 to a low level, assuming that there is no possibility that the confidential information 1 is included in the input voice at this time.

In this example, the confidential information phoneme sequence starting with the phoneme sequence "den" is not registered in the confidential information list 141. Therefore, it is assumed that the determination unit 13 also determines that the confidential information 2 to 10 do not have a prefix match with the output phoneme "den" as in the confidential information 1. As a result, the confidential information determination signal corresponding to the confidential information 2 to 10 is also set to a low level. The determination unit 13 takes a logical sum for the level of the confidential information determination signal (individual) corresponding to the confidential information 1 to 10, and outputs the confidential information determination signal (comprehensive) indicating the result of the logical sum to the voice data concealment unit 15. .. At time t0, the determination unit 13 determines the confidential information whose signal level is set to a low level based on the fact that all the confidential information determination signals (individual) corresponding to the confidential information 1 to 10 are set to a low level. The signal (comprehensive) is output to the voice data concealment unit 15. Further, the determination unit 13 determines that none of the confidential information 1 to 10 completely matches the output phoneme "den", and sets the perfect match signal (individual) corresponding to the confidential information 1 to 10 at a low level. .. The determination unit 13 takes the logical sum of the confidential information 1 to 10 and outputs a low-level perfect match signal (comprehensive) to the voice data concealment unit 15.

When the voice data concealment unit 15 receives the concealment information determination result from the determination unit 13 (step S21), the voice concealment process shown in FIG. 2 is started. At time t0, the signal levels of the confidential information determination signals corresponding to the confidential information 1 to 10 are all low levels. Therefore, the output control unit 150 determines that the digital voice data included in the section 1 currently being processed (that is, at time t0) can be output from the buffer 151 (YES in step S22), and the section 1 Digital voice data (corresponding to the output phoneme “den”) is output from the buffer 151 (step S26), and the voice concealment process is terminated. As a result, the digital voice data corresponding to the output sound that does not match the confidential information sound element sequence, that is, the digital voice data that does not correspond to a part of the secret information, is immediately output in the voice concealment processing after the confidential information determination processing, and is output to the network. It is promptly transmitted to the voice recognition server 100 via 3.

As shown in FIG. 3, after the lapse of time t0, the digital voice data (input voice “wa”, “bang”, “go”) included in the sections 2 to 5 recognized by the voice recognition unit 12 by the time t1. Similarly, the determination unit 13 determines that any of the confidential information phoneme sequences in the confidential information list 141 is prefix-matched or does not completely match, and is output from the buffer 151 by the output control unit 150. Therefore, during the period from time t0 to t1, digital voice data for five sections (sections 1 to 5) corresponding to the input voice "Denwabangowa" is transmitted to the voice recognition server 100. Further, as shown in FIG. 3, the output digital voice data does not remain in the buffer 151.

As shown in FIG. 3, at time t2, the voice recognition unit 12 and the buffer 151 output digital voice data (corresponding to the input voice “zero”) included in the section 6 from the voice input device 11, and the voice recognition unit 12 Is output to the determination unit 13 as an output phoneme corresponding to the Japanese reading of "zero". At time t2, the determination unit 13 executes the confidential information determination process for the output phoneme “zero” corresponding to the digital voice data included in the section 6. As described above, in this example, the confidential information phoneme sequence corresponding to the confidential information 1 is "Zero Cue Zero Ichini Sanyon Gourokunana". Since the output phoneme "zero" and the first phoneme "zero" in the secret information phoneme sequence of the secret information 1 are the same in the determination unit 13, the input output phoneme and the secret information phoneme series of the secret information 1 are prefix-matched. Then, the confidential information determination signal (individual) corresponding to the confidential information 1 is set to a high level. In addition, the determination unit 13 holds an output phoneme that previously matches the confidential information phoneme sequence in a predetermined storage area.

Further, in this example, the confidential information phoneme sequence starting with the phoneme sequence "zero" is not registered in the confidential information list 141 other than the confidential information 1. Therefore, the determination unit 13 determines that the confidential information phoneme sequence and the output phoneme "zero" do not match in front of the confidential information 2 to 10, and lowers the confidential information determination signal (individual) corresponding to the confidential information 2 to 10. Set to (Low) level. The determination unit 13 takes the logical sum of the confidential information 1 to 10 and outputs a high level confidential information determination signal (comprehensive) to the voice data confidential unit 15. Further, the determination unit 13 determines that although the confidential information determination signal (comprehensive) is at a high level, none of the confidential information 1 to 10 completely matches the output phoneme "zero", and the confidential information 1 to 10 are used. Set the corresponding exact match signal (individual) to a low level. The determination unit 13 takes the logical sum of the confidential information 1 to 10 and outputs a low-level perfect match signal (comprehensive) to the voice data concealment unit 15.

The output control unit 150 of the voice data concealment unit 15 confirms the signal level of the input concealment information determination signal (comprehensive) in the voice concealment processing executed based on the concealment information determination processing at time t2, and the buffer output is output. It is determined whether or not it is possible (step S22). Since the confidential information determination signal (individual) corresponding to the confidential information 1 is at a high level at time t2, the confidential information determination signal (comprehensive) is also set to a high level. Therefore, at time t2, the output control unit 150 determines that buffer output is impossible (NO in step S22), and stops the output of the digital voice data (section 6 in this example) stored in the buffer 151. (Step S23). As a result, the voice concealment device 10 according to the present embodiment determines that a part of the input voice (digital voice data in each section) coincides with the concealed information in the front, that is, the input voice may include the concealed information. At the stage, the content of the input voice can be concealed. Therefore, in the voice concealment device 10, the user conceals at an earlier stage than determining whether or not the content contained in the input voice completely matches the keyword such as the concealed information registered in advance as in the conventional case. It is possible to prevent the leakage of desired confidential information.

As shown in FIG. 3, at time t3, the voice recognition unit 12 and the buffer 151 output digital voice data (corresponding to the input voice “cue”) included in the section 7 from the voice input device 11, and the voice recognition unit 12 Outputs the output phoneme "kyuu", which corresponds to the Japanese reading, to the determination unit 13. At time t3, the determination unit 13 combines the output phoneme "queue" corresponding to the digital voice data included in the section 7 and the output phoneme "zero" stored at time t2 to form a combined output phoneme sequence called "zero queue". It is generated and the confidential information determination process is executed for this combined output phoneme sequence. In this way, the determination unit 13 determines the output phoneme corresponding to the subsequent section (in this example, the section 7 or later) corresponding to the section (section 6 in this example) that has been prefix-matched with the secret information phoneme series. Combine in chronological order to generate a combined output phoneme sequence. The determination unit 13 generates a combined output phoneme sequence, and repeats the determination of the prefix match with the secret information phoneme sequence for the generated combined output phoneme sequence, so that the series of input voices and the secret information are finally completely matched. It can be determined whether or not.

At time t3, the determination unit 13 determines that the combined output phoneme sequence "zero queue" and the phoneme "zero queue" continuous from the beginning of the secret information phoneme series of the secret information 1 coincide with each other, and determines the secret information corresponding to the secret information 1. Set the signal (individual) to a high level. Further, in this example, the confidential information phoneme sequence starting with the phoneme sequence "zero queue" is not registered in the confidential information list 141 other than the confidential information 1.

Further, the determination unit 13 determines whether or not the output phoneme "queue" input at time t2 alone matches the secret information phoneme sequence of the secret information 1 to 10 in advance. The determination unit 13 secretly informs about the output phoneme constituting the end of the combined output phoneme sequence, that is, the latest output phoneme, regardless of whether or not there is a secret information phoneme sequence that matches the front of the combined output phoneme sequence. Judge the prefix match with the phoneme sequence. As a result, the voice concealment device 10 according to the present embodiment always extracts a concealed information phoneme sequence that matches the latest output phoneme in advance, and determines whether or not the input voice may match a keyword (confidential information) including personal information and the like. It can be judged properly.

In this example, it is assumed that the confidential information phoneme sequence starting with the phoneme "cue" is not registered in the confidential information list 141. Therefore, the determination unit 13 sets the confidential information determination signal (individual) corresponding to the confidential information 2 to 10 to a low level. The determination unit 13 takes the logical sum of the confidential information determination signals (individual) corresponding to each of the confidential information 1 to 10 and outputs a high level confidential information determination signal (comprehensive) to the voice data confidential unit 15. .. Further, the determination unit 13 determines that although the confidential information determination signal (comprehensive) is at a high level, none of the confidential information 1 to 10 completely matches the combined output phoneme series "zero queue" and the output phoneme "queue". Then, the exact match signal (individual) corresponding to the confidential information 1 to 10 is set to a low level. The determination unit 13 takes a logical sum of the exact match signals (individual) corresponding to each of the confidential information 1 to 10 and outputs a low-level perfect match signal (comprehensive) to the voice data concealment unit 15. The exact match signal set at a low level is output to the voice data concealment unit 15.

The output control unit 150 of the voice data concealment unit 15 stops the output of the digital signal data from the buffer 151 in the voice concealment process executed based on the concealment information determination process at the time t3 (step S22). Flow from NO to step S23). As a result, as shown in FIG. 3, at the end of the voice concealment processing based on the concealment information determination processing at time t3, the buffer 151 contains the digital signal data (input voice “zero”” and “input voice“ zero ”” included in the section 6 and the section 7. Equivalent to "cue") is accumulated.

As shown in FIG. 3, the digital phonemes included in the sections 8 to 16 output to the voice recognition unit 12 during the period from the time t3 to the time t4 are the output phonemes "zero" and "ichi (", respectively. It is recognized as "Ichi", "nii", "san", "yon", "go", and "roku", and is sequentially recognized by the determination unit 13 as the combined output phoneme series "zero queue". "Zero" (combination of output phonemes corresponding to sections 6 to 8), "Zero cue zero zero" (combination of output phonemes corresponding to sections 6 to 9), "Zero cue zero zeroichi" (output corresponding to sections 6 to 10) Phoneme combination), "Zero Cue Zero Zero Ichinie" (combination of output phonemes corresponding to sections 6 to 11), "Zero Cue Zero Zero Ichinie Sun" (combination of output phonemes corresponding to sections 6 to 12), "Zero Cue" "Zero Zero Ichinie Sanyon" (combination of output phonemes corresponding to sections 6 to 13), "Zero Cue Zero Zero Ichinie Sanyongo" (combination of output phonemes corresponding to sections 6 to 14), "Zero Cue Zero Zero Ichinie Sanyon Goroku" "(Combining of output phonemes corresponding to sections 6 to 15) is generated.

All of these combined output phoneme sequences generated in the period from the lapse of time t3 to time t4 coincide with the secret information phoneme sequence corresponding to the secret information 1. Therefore, the confidential information determination signal (individual) corresponding to the confidential information 1 is set to a high level during the period from the lapse of the time t3 to the time t4, and the confidential information determination signal (individual) corresponding to the confidential information 2 to 10. ) Is at a low (Low) level, the confidential information determination signal (comprehensive) set to a high level by the logical sum is output to the voice data concealment unit 15. Based on this, the output control unit 150 continues to stop the output of the digital audio data from the buffer 151 from the time t3 lapse to the time t4. Therefore, as shown in FIG. 3, at the end of the voice concealment processing based on the execution of the personal determination processing at time t4, the buffer 151 stores digital voice data for 10 sections of sections 6 to 15. To. In this example, the output control unit 150 outputs the digital voice data during the period in which the digital voice data (corresponding to the input voice "Zero Cue Zero Zero Ichini Sanyon Goroku") included in the sections 6 to 15 is stored in the buffer 151. Is the output stop period when is stopped.

Further, in the voice concealment device 10 according to the present embodiment, the determination unit 13 determines that the concealed information phoneme sequence and the output phoneme (or the combined output phoneme sequence) are in front of each other, so that the output phoneme and the concealed information phoneme sequence are matched. It is possible to determine the degree, that is, the degree of matching between a series of input voices and a keyword including personal information and the like. In this example, the determination unit 13 has a higher degree of matching between the output phoneme and the secret information phoneme sequence as the ratio of the phonemes that are in front of the combined output phoneme sequence among the phonemes constituting the secret information phoneme sequence is larger. Is determined. For example, the degree of matching with the secret information phoneme series "Zero Cue Zero Ichini Sanyon Gourokunana" corresponding to the secret information 1 is higher than that of the combined output phoneme series "Zero Cue Zero" (combination of sections 6 to 8). The series "zero queue zero zero" (combination of sections 6-9) is higher. Therefore, the combined output phoneme sequence that completely matches the confidential information phoneme sequence has the highest degree of matching (100%). As described above, the higher the degree of matching between the output phoneme and the confidential information phoneme sequence, the greater the risk (risk) that the keyword is included in the series of input voices.
Further, when there is no possibility that the keyword is included in the series of input voices (the input voice and the keyword match), the degree of matching between the series of input voices and the keyword including personal information is 0%. When the determination unit 13 determines in the voice concealment device 10 that the output phoneme and the concealment information phoneme sequence are in front of each other, that is, when the concealment information determination signal (comprehensive) is set to a high level. Therefore, the degree of matching between the series of input voices and the keywords including personal information and the like is a value larger than 0%. In this example, the degree of matching between the output phoneme and the phoneme sequence is at time t2 when it is determined that the output phoneme "zero" (corresponding to the digital voice data included in the section 6) and the confidential information phoneme sequence of the confidential information 1 are prefix-matched. The value is larger than 0%. When the degree of matching is a value larger than 0%, there is a possibility that a keyword is included in a series of input voices.

As shown in FIG. 3, at time t5, digital voice data (corresponding to the input voice “Nana”) included in the section 12 is output from the voice input device 11 to the voice recognition unit 12 and the buffer 151, and the voice recognition unit 12 Therefore, the output phoneme "nana" corresponding to the Japanese reading of "Nana" is output to the determination unit 13. At time t5, the determination unit 13 determines the output phoneme "nana" corresponding to the digital voice data included in the section 16 at the end of the combined output phoneme series generated by the output phonemes corresponding to the digital voice data included in the sections 6 to 15. (Nana) ”is combined to generate a combined output phoneme sequence“ Zero Cue Zero Zero Ichinigo ”(combination of sections 6 to 16).

The combined output phoneme sequence "Zero Cue Zero Zero Ichini Sanyon Gorokunana" generated by combining the sections 6 to 16 is prefix-matched and completely coincides with the secret information phoneme sequence corresponding to the secret information 1. Therefore, the determination unit 13 sets the confidential information determination signal (comprehensive) to a high level, and the degree of matching between the combined output phoneme sequence and the confidential information phoneme sequence is the highest (perfect match). The signal level of the exact match signal (individual) corresponding to the confidential information 1 is set to a high level based on the above. Further, the determination unit 13 takes a logical sum of the exact match signals (individual) corresponding to the confidential information 1 to 10 and outputs a high-level perfect match signal (comprehensive) to the voice data concealment unit 15.

The output control unit 150 stops the output of the digital signal data from the buffer 151 based on the high level signal level of the confidential information determination signal (total) in the voice concealment processing at time t5. (Flow from NO in step S22 to step S23). Further, the output control unit 150 determines that the signal level of the perfect match determination signal is high and the input voice contains content that completely matches the keyword such as personal information (YES in step S24). The digital signal data currently stored in the buffer 151 is deleted (step S25). As a result, the digital voice data corresponding to the sections 6 to 16 stored in the buffer 151, that is, the digital voice data whose contents completely match the confidential information 1 is deleted from the buffer 151. As described above, the voice concealment device 10 according to the present embodiment can delete digital audio data having a content that completely matches any of the keywords (confidential information) such as personal information registered in the confidential information list 141 from the buffer 151. .. Therefore, in the voice concealment device 10, keywords such as personal information are not transmitted to the voice recognition server 100, and the voice concealment device 10 can surely prevent the confidential information registered in advance by the user from leaking. can.

Next, another example of the flow of the confidential information determination process in the determination unit 13 will be described with reference to FIGS. 1 to 3. FIG. 4 is a diagram illustrating another example of the flow of the confidential information determination process in chronological order. FIG. 4 shows the confidential information determination process in the case where the voice input device 11 inputs a prefix matching voice including a content that matches the confidential information 1 (for example, the telephone number “090012345667”) in the prefix but does not completely match. An example is shown. It is assumed that the prefix matching voice of this example is a series of voices such as "phone number is 09001254678". Since the items shown in the first to eleventh columns from the top of FIG. 4 are the same as the items shown in the first to eleventh columns from the top of FIG. 3, the description thereof will be omitted.

As shown in FIG. 4, the prefix matching voices of this example are "den", "wa", "ban", "go", "wa", "zero", "cue", "zero", "zero", "ichi", "knee", and "go". It is output to the voice recognition unit 12 in the order of sections 101 to 116 (in chronological order) as 16 sections including digital voice data corresponding to the contents of "Yon", "Roku", "Nana", and "Hachi" respectively. ..

As shown in FIG. 4, the determination unit 13 has output phonemes "den" and "wa" corresponding to sections 101 to 105 in which the voice recognition unit 12 has sequentially performed voice recognition processing in the period from time t10 to time t11. Confidential information determination processing is executed for "wa", "ban", "gou", and "wa". The determination unit 13 determines that none of the output phonemes corresponding to the sections 101 to 105 is a prefix match or an exact match with any of the secret information phoneme sequences registered in the secret information list 141. Therefore, in the voice concealment processing during the period from time t10 to t11 shown in FIG. 4, the output control unit 150 has five sections (sections) called "Denwabangowa" as in the case of time t0 to time t1 shown in FIG. The digital audio data of 101 to 105) is output from the buffer 151 in chronological order (flow from YES in step S22 to step S26).

Further, in the confidential information determination process at time t12, the determination unit 13 determines that the output sound element "zero" corresponding to the digital voice data included in the section 106 is forward-matched with the confidential information element sequence of the confidential information 1. The determination is made, and the confidential information determination signal (comprehensive) having a high signal level is output to the voice data concealment unit 15. Based on this, in the voice concealment processing at the time t12, the output control unit 150 stops the output of the digital signal data from the buffer 151 (from NO in step S22 to step S23), similarly to the time t2 shown in FIG. flow). After that, in the period from time t13 to time t14, the determination unit 13 generates a combined output phoneme series in which the sections 106 to 111 are sequentially combined, and the generated combined output phoneme series and the secret information phoneme series of the secret information 1 are prefix-matched. Then it is determined.

Therefore, the confidential information determination signal (total) is set to a high level during the period from the time t13 to the time t14, and the output control unit 150 continues to stop the output of the digital voice data from the buffer 151. Therefore, as shown in FIG. 4, in this example, the period from time t13 to time t14 is the output stop period during which the output of digital audio data from the buffer 151 is stopped, and the digital audio for six sections from section 106 to section 111 is stopped. Data (corresponding to "zero cue zero zero ichiny" in the prefix matching voice (input voice) shown in FIG. 4) is stored in the buffer 151.

As shown in FIG. 4, at time t15, digital voice data (corresponding to the input voice “go”) included in the section 112 is output from the voice input device 11 to the voice recognition unit 12 and the buffer 151, and the voice recognition unit 12 Therefore, the output phoneme "go" corresponding to the Japanese reading of "go" is output to the determination unit 13. At time t15, the determination unit 13 determines the output phoneme "go" corresponding to the digital voice data included in the section 12 at the end of the combined output phoneme series generated by the output phonemes corresponding to the digital voice data included in the sections 106 to 111. Is combined to generate a combined output phoneme sequence "Zero Cue Zero Zero Ichinigo" (combination of output phonemes corresponding to sections 106 to 112).

The combined output phoneme sequence generated by the determination unit 13 at time t15 does not coincide with the corresponding secret information phoneme series "Zero Cue Zero Zero Ichini Sanyon Gorokunana" of the secret information 1. Further, it is assumed that the confidential information phoneme sequence corresponding to the confidential information 2 to 10 does not match forward. Further, the determination unit 13 determines the prefix match with the secret information phoneme sequence corresponding to the secret information 1 to 10 only by the output phoneme “go”. In this example, it is assumed that the confidential information phoneme sequence starting with the phoneme sequence "go" is not registered in the confidential information list 141. Therefore, the determination unit 13 sets all the confidential information determination signals (individual) corresponding to each of the confidential information 1 to 10 to a low (Low) level, and is low by the logical sum of the signal levels of the confidential information determination signals (individual). The confidential information determination signal (comprehensive) set to the level is transmitted to the voice data concealment unit 15. In this way, the determination unit 13 determines the output phoneme that constitutes the end of the combined output phoneme sequence, that is, the latest output phoneme, even when the secret information phoneme sequence that has previously matched the combined output phoneme sequence no longer exists. Is determined to be a prefix match with the confidential information phoneme sequence. As a result, it is possible to appropriately determine whether or not the input voice may match the keyword including personal information and the like.

In the voice concealment processing executed based on the concealment information determination processing at time t15, the output control unit 150 of the voice data concealment unit 15 has a low signal level of the concealment information determination signal (comprehensive). Based on this, it is determined that buffer output is possible (YES in step S22), and the digital audio data stored in the buffer 151 is output (step S26). As shown in FIG. 4, at time t15, the buffer 151 stores digital voice data for seven sections corresponding to sections 6 to 12 (corresponding to the input voice “zero cue zero zero ichinigo”). As described above, digital audio data is input / output to / from the buffer 151 by the first-in first-out method. Therefore, the output control unit 150 sequentially outputs digital audio data in chronological order, that is, in order from the section previously stored in the buffer 151. In this example, the sections 106 to 112 are continuously output from the buffer 151 in this order. As a result, the digital voice data included in the plurality of sections is transmitted to the voice recognition server 100 via the network 3 while maintaining the continuity of the series of input voices in the time series. The voice recognition server 100 performs voice recognition on the received digital voice data and transmits the voice recognition result to the server 500. Therefore, the server 500 can return the processing result (for example, the information retrieval result) according to the content spoken by the user to the voice processing device 1.

In the present invention, the input / output method of digital audio data in the buffer 151 is not limited to the first-in first-out method. If the voice concealment device 10 can transmit the digital voice data for a plurality of sections stored in the buffer 151 to the voice recognition server 100 while maintaining the time-series order, the digital voice in the buffer 151 is used in any manner. Data may be input / output.

As shown in FIG. 4, the digital voice data included in the sections 113 to 116 output to the voice recognition unit 12 during the period from the time t15 to the time t16 are the output phonemes "yon" and "roku", respectively. (Roku) ”,“ nana ”, and“ hachi ”are recognized, and the determination unit 13 sequentially executes the confidential information determination process. The output phonemes corresponding to the digital voice data included in the sections 113 to 116 are the secret information stored in the secret information list 141 in the determination unit 13 in the same manner as the output phonemes corresponding to the digital voice data included in the sections 101 to 105. It is assumed that it is determined that there is no prefix match or perfect match with any of the confidential information phoneme sequences 1 to 10. Therefore, the digital voice data included in the sections 113 to 116 is immediately output from the buffer 151 by the output control unit 150 after the confidential information determination processing. Therefore, in the period from the lapse of the time t15 to the time t16, the digital voice data for four sections of "Yon", "Roku", "Nana", and "Hachi" are transmitted to the voice recognition server.

As described above with reference to FIGS. 1 to 4, the voice concealment device 10 according to the present embodiment can store a concealed information phone series (an example of phone series information) in which a plurality of keywords are converted into sound information. Output voice (output sound element information) which is information recognized at the sound element level by sequential voice recognition processing of input voice (input voice data) from the storage unit 14 and a predetermined voice generation source (for example, the user of the voice processing device 1). The voice recognition unit 12 output as an example) and the secret information sound element sequence (an example of the stored sound element information) in which the output sound element output by the voice recognition unit 12 is stored in the secret information list 141 of the storage unit 14 are the same. Corresponds to the determination unit 13 capable of sequentially determining whether or not the input voice data and the keyword may match based on the degree, and the input voice data determined by the determination unit 13 to match the keyword. It is provided with an audio data concealment unit 15 capable of concealing the contents of digital audio data.
Further, the determination unit 13 has input voice data and keywords from a predetermined voice source based on the degree of prefix matching between the output phoneme output by the voice recognition unit 12 and the confidential information phoneme sequence stored in the storage unit 14. It is possible to sequentially determine whether or not there is a possibility of matching with (personal information, etc.), and the voice data concealment unit 15 determines that the degree of prefix matching with the concealment information phoneme sequence is equal to or higher than a predetermined degree by the determination unit 13. The content of the voice data corresponding to the determined output phoneme can be concealed.

As a result, in the voice concealment device 10 according to the present embodiment, for example, a part of the input voice may be prefix-matched with a keyword such as personal information (information that the user wants to conceal (confidential information)), and the input voice may match the keyword. The content of the input voice can be concealed at the stage when the sex occurs. Therefore, the voice concealment device 10 reliably and promptly performs an individual at an early stage in voice recognition of a series of input voices (for example, "phone number is 090012345667"), that is, at a stage where there is a possibility of matching without waiting for an exact match. It is possible to prevent the leakage of confidential information including information.

Further, in the voice concealment device 10 according to the present invention, in the determination unit 13, the degree of matching between the output phoneme which is the recognition result of the input voice by the voice recognition unit 12 and the concealed information phoneme sequence stored in the concealed information list 141 ( For example, whether or not there is a prefix match) is determined. Both output phonemes and confidential information phoneme sequences are sound information. For this reason, the voice concealment device 10 needs to perform a process of matching the data type of the concealed information (for example, text) registered in advance with the data type of the input voice each time the concealment device 10 determines the match with the concealed information. not. Therefore, the voice concealment device 10 reduces the processing load at the time of concealment information determination processing and maintains the rapid processing speed to compare the content of the input voice with the concealed information (for example, a keyword including personal information). It can be done efficiently.

Further, in the voice concealment device 10 according to the present invention, in the voice data concealment unit 15, the input voice data from a predetermined voice generation source (for example, the user of the voice processing device 1) is completely combined with a keyword such as personal information by the determination unit 13. When it is determined that there is no possibility of matching, the content of the digital voice data corresponding to the input voice is not concealed. As a result, the voice concealment device 10 can quickly output the digital voice data and transmit it to the voice recognition server 100 when it is determined that there is no possibility that the input voice data contains the concealed information. Therefore, the voice concealment device 10 can both prevent the leakage of information (information to be concealed) desired to be concealed by the user of the voice processing device 1 and maintain the service provision speed using the voice recognition server 100. ..

(1-5) Voice Concealment Program The voice concealment device 10 of the present embodiment is a software program (voice concealment program) developed in the voice processing device 1 which is an information terminal device such as a computer or a smart speaker, and is a software program. The CPU of the voice processing device 1 executes arithmetic processing according to the command of. That is, in the voice concealment program according to the present embodiment, a predetermined information terminal device (voice processing device 1 in this example) is stored in a storage unit 14 capable of storing a concealed information sound element sequence in which a plurality of keywords are converted into sound element sequence information. The voice recognition unit 12 and the voice recognition unit 12 output the input voice input from a predetermined voice generation source (for example, the user of the voice processing device 1) as an output sound recognized at the sound element level by the sequential voice recognition process. Judgment that it is possible to sequentially determine whether or not there is a possibility that the input voice data and the keyword match based on the degree of matching between the output voice and the secret information sound element sequence information stored in the secret information list 141 of the storage unit 14. The content of the digital voice data corresponding to the input voice data determined by the unit 13 and the determination unit 13 to match the keyword such as personal information is made to function as the voice data concealment unit 15.

(Modification 1 of the first embodiment)
Next, a modification 1 of the first embodiment will be described with reference to FIG. 5A. In the first modification, the contents of the confidential information determination process and the voice concealment process executed by the voice concealment device 10 are different from those of the first embodiment. Specifically, in the first modification, the voice concealment device 10 is based on whether or not the input voice data includes information (confidential prediction information) that is predicted to speak content including personal information and the like. It is determined whether or not there is a possibility that confidential information such as personal information is included in the input audio data, and whether or not the buffer output of the digital audio data is possible is determined according to the determination. Here, as the confidential prediction information, keywords such as "telephone number is" and "name is" are assumed in which personal information is predicted to be uttered in succession.

In the first modification, the storage unit 14 stores a secret prediction information list. Like the secret information list 141, the secret prediction information list stores a plurality of secret prediction information in a table format. Further, the confidential prediction information stored in the confidential prediction information list is information registered in advance by the user of the voice processing device 1 using a predetermined information terminal device connected to the voice processing device 1, similar to the confidential information. be. In the first modification, the storage unit 14 may store only the secret prediction information list, or may store both the secret prediction information list and the secret information list 141.

FIG. 5A is a diagram illustrating the confidential information determination process and the voice confidentiality process in the first modification in chronological order. In this example, in the concealment prediction information list, the concealment prediction phoneme series "(Denwabangowa)" (confidential prediction information 1) and "name is" indicating the content of "telephone number is" as the concealment prediction information. It is assumed that at least the phoneme series "(Namaeva)" (confidential prediction information 2) is accumulated. FIG. 5A shows an example of confidential information determination processing when a voice (in this example, “telephone number is 090012345667”) including a content that completely matches the confidential prediction information 1 is input from the voice input device 11 as input voice. ing. The series of input voices in this example has the same contents as the input voices in FIG. Therefore, in FIG. 5A, as in FIG. 3, the input voice is divided into 16 sections of sections 1 to 16.

As shown in FIG. 5A, at time t20, the determination unit 13 forward-matches the output phoneme “den” corresponding to the digital voice data included in the section 1 with the secret prediction information phoneme sequence of the secret prediction information 1. Then it is determined. Further, the determination unit 13 holds the output phonemes that are in front of the concealed predicted phoneme sequence in a predetermined storage area, and the combined output phonemes in which the digital voice data corresponding to the sections 2 to 5 are sequentially combined in the period up to the time t21. Generate a series.

The determination unit 13 has a combined output phoneme sequence "Denwabangowa" (combination of output phonemes corresponding to intervals 1 to 5) generated in the period t21 and a secret predicted phoneme sequence ("Denwabangowa") of the secret prediction information 1. It is determined that the match is perfect, and the secret prediction information determination signal (individual) corresponding to the secret prediction information 1 is set to a high level. The concealment prediction information determination signal having a high signal level indicates that the output phoneme completely matches the concealment prediction phoneme sequence. Further, the determination unit 13 determines at time t21 that the combined output phoneme sequence “Denwabangowa” and the secret prediction phoneme sequence (“Namaeva”) of the secret prediction information 2 do not completely match, and corresponds to the secret prediction information 2. The confidential prediction information judgment signal (individual) is set to the low level. The concealment prediction information determination signal (individual) having a low signal level indicates that the output phoneme does not completely match the concealment prediction phoneme sequence. The determination unit 13 takes the logical sum of the confidential prediction information determination signals (individual) corresponding to the confidential prediction information 1 and 2, and outputs a high-level confidential prediction information determination signal (comprehensive) to the voice data concealment unit 15.

Further, when the secret prediction information determination signal (individual) corresponding to the secret prediction information 1 is set to a high level at the time t21, the determination unit 13 sets a predetermined fixed period (for example, 5 seconds) after the time t22. Each time the output sound element is output from the voice recognition unit 12, the concealment prediction information determination signal (individual) corresponding to the concealment prediction information 1 is set to a high level, and the concealment prediction information determination signal (comprehensive) at a high level is set. Is output to the voice data concealment unit 15. When the output control unit 150 of the voice data concealment unit 15 receives the high-level concealment prediction information determination signal from the determination unit 13 at time t21, after the time t22 until the low-level concealment prediction information determination signal (comprehensive) is received. The digital audio data input to the buffer 151 from the audio input device 11 is deleted from the audio input device 11 over the above-mentioned fixed period. That is, in the first modification, this fixed period is the output stop period during which the digital audio data is not output from the buffer 151. In this example, as shown in FIG. 5A, during the period from time t22 to time t23, the determination unit 13 outputs the level concealment prediction information determination signal (comprehensive) to the voice data concealment unit 15, and buffers during the period. The digital audio data included in the sections 6 to 16 input to 151 (corresponding to the input audio "Zero Cue Zero Zero Ichiny Sanyon Gorokunana") is sequentially deleted by the output control unit 150.

As described above, in the first modification, the voice concealment device 10 is determined by the determination unit 13 to include the content that matches the concealment prediction information in the input voice, that is, the input voice data and the keyword (voice concealment information). Digital voice data corresponding to the input voice data input from the user of the voice processing device 1 during a predetermined fixed period (output stop period) after the judgment based on the judgment that there is a possibility of matching with. The contents of (digital voice data input from the voice input device 11 to the buffer 151) can be concealed (buffer output is stopped). As a result, the voice concealment device 10 in the first modification is the content of the input voice at a time before the user speaks the confidential information including personal information, that is, at a time when it is predicted that the user will speak the confidential information. Can be kept secret. Further, during the output stop period, the output control unit 150 deletes the digital voice data input to the buffer 151. As a result, the voice concealment device 10 in the first modification can surely prevent the leakage of information (confidential information) that the user desires to conceal, such as personal information.

(Modification 2 of the first embodiment)
Next, a modification 2 of the first embodiment will be described with reference to FIG. 5B. In the modified example 2, the method of determining the output stop period is different from that of the modified example 1 when it is determined by the determination unit 13 that the input voice contains the content that matches the confidentiality prediction information. FIG. 5B is different from FIG. 5A in that the input voice is not divided into a plurality of sections and is associated with the voice section, but is the same as FIG. 5A except for this point. In the second modification, the engine unit 120 of the voice recognition unit 12 has a voice section in which the user speaks continuously for a predetermined time, and a fixed time (for example, 500 milliseconds) or more caused by breathing during the speech. It determines a silent section (a section in which no voice exists), recognizes digital voice data at the phoneme level for each voice section, and outputs the digital voice data to the determination unit 13.

As shown in FIG. 5B, in this example, the engine unit 120 recognizes the digital voice data included in the voice section ac1 (time t30 to time t31) corresponding to a series of input voices called "Denwabangowa" at the phoneme level, and recognizes the time. At t31, the output phoneme "denwabangouwa" is output to the determination unit 13. The determination unit 13 determines that the output phoneme completely matches the secret prediction phoneme sequence (“Denwabangowa”) of the secret prediction information 1 at time t31, and sets the secret prediction information determination signal (individual) corresponding to the secret prediction information 1 high (individual). High) Set to level. Further, at time t31, the determination unit 13 determines that the output phoneme and the secret prediction phoneme sequence (“Namaeva”) of the secret prediction information 2 do not completely match, and the secret prediction information determination signal (individual) corresponding to the secret prediction information 2. ) Is set to the low level. The determination unit 13 takes a logical sum of the concealment prediction information determination signals (individual) corresponding to the concealment prediction information 1 and 2, and outputs a high-level concealment prediction information determination signal (comprehensive) to the voice data concealment unit 15.

Further, when the determination unit 13 sets the confidential prediction information determination signal (individual) corresponding to the confidential prediction information 1 to a high level at time t31, the voice section is subsequently set to a predetermined number of minutes (for example, two). In the corresponding period, the secret prediction information determination signal (individual) corresponding to the secret prediction information 1 is set to a high level, and the output sound corresponding to the digital voice data included in each voice section is output from the voice recognition unit 12. A high-level concealment prediction information determination signal (comprehensive) is output to the voice data concealment unit 15 each time. When the output control unit 150 of the voice data concealment unit 15 receives a high-level concealment prediction information determination signal from the determination unit 13 at time t31, the output control unit 150 corresponds to a predetermined number of minutes (for example, two) of voice sections after time t32. Delete the digital audio data from the buffer 151. In this example, as shown in FIG. 5B, the subsequent voice section ac2 (corresponding to the period from time t32 to time t33) including the digital voice data having the content matching the secret prediction phoneme sequence of the secret prediction information 1. , Ac3 (corresponding to the section from time t34 to time t35), the digital voice data (corresponding to the input voice "Zero Cue Zero Zero Ichini Sanyon Gorokunana") is deleted from the buffer 151. As described above, in the modification 2, the period corresponding to the predetermined number of voice sections (2 in this example) is the output stop period. Further, after the end of the output stop period, the determination unit 13 sets the confidential prediction information determination signal (individual) corresponding to the confidential prediction information 1 to a low level, and sets the low level confidential prediction information determination signal (comprehensive). Is output to the voice data concealment unit 15. As a result, after the end of the output stop period, the output control unit 150 can output the digital audio data input to the buffer 151 to the buffer.

As described above, in the present modification 2, the voice concealment device 10 is determined by the determination unit 13 that the input voice data includes the content that matches the concealment prediction information, and after the determination, the predetermined voice generation source is determined. Of the digital audio data corresponding to the input audio data input from, the content of the audio data in a predetermined number of audio sections can be concealed. That is, the output control unit 150 conceals the digital voice data input from the voice input device 11 to the buffer 151 during the period corresponding to a predetermined number of voice sections (output stop period) after the judgment by the judgment unit 13 (buffer output). Can be stopped). As a result, the voice concealment device 10 in the second modification can conceal the content of the input voice when it is predicted that the user will utter the confidential information. Further, during the output stop period, the output control unit 150 deletes the digital voice data input to the buffer 151. As a result, the voice concealment device 10 in the second modification can surely prevent the leakage of information (confidential information) that the user desires to conceal, such as personal information.

The voice concealment device 10 can execute both the modified example 1 in which the output stop period is a predetermined fixed time and the modified example 2 in which the period corresponding to a predetermined number of voice sections is the output stop period. May be good. For example, the confidential information determination process and the voice concealment process in the modified example 1 and the confidential information determination process and the voice concealment process in the modified example 2 may be switched or executed under predetermined conditions. For example, the voice concealment device 10 may be configured such that a predetermined fixed time is set as an output stop period, or a period corresponding to a predetermined number of voice sections is set as an output stop period. For example, the voice concealment device 10 is configured such that a period belonging to at least one of a predetermined fixed time and a period corresponding to a predetermined number of voice sections is set as an output stop period. May be good.

In the present embodiment, the voice concealment device 10 determines whether or not there is a possibility that the input voice and the keyword match, based on the prefix matching between the concealed information phoneme sequence and the output phoneme, but the concealed information phoneme sequence. Of these, the part to be the target of prefix matching (the number from the beginning of the secret information phoneme series) may be set for each secret information phoneme series. As a result, the possibility of matching can be more appropriately determined according to the content of the confidential information, the characteristics of the voice recognition unit 12, and the like. Further, the calculation of the degree of coincidence is not limited to the prefix match, and may be calculated by, for example, an intermediate match (match with the portion of the front part excluding the phoneme at the beginning which is likely to be confused). The voice concealment device 10 determines the possibility of matching based on the degree of matching, so that personal information and the like can be reliably and promptly determined at an early stage in voice recognition, that is, at a stage where there is a risk of matching without waiting for an exact match. It is possible to prevent the leakage of confidential information including.

2. 2. Second Embodiment Next, the voice concealment device according to the second embodiment of the present invention will be described with reference to FIGS. 6 and 7. The voice concealment device 20 according to the present embodiment is different from the voice concealment device 10 according to the first embodiment in that it is configured independently rather than as a component of an information terminal device or the like. The voice concealment device 20 is, for example, a device that is arranged in the vicinity of an information terminal device having a voice recognition function such as a smart speaker, and conceals the concealed information that the user of the information terminal device desires to conceal by outputting a disturbing sound. ..

As shown in FIG. 6, the voice concealment device 20 includes a voice input unit 21, an echo canceller 22, a voice recognition unit 12, a determination unit 13, a storage unit 14, and a voice data concealment unit 25. There is. In the voice concealment device 20, the configuration having the same operation and effect as the voice concealment device 10 according to the first embodiment is designated by the same reference numerals as the voice concealment device 10 shown in FIG. 1, and detailed description thereof will be omitted.

The voice input unit 21 has the same function as the voice input device 11 included in the voice processing device 1 (see FIG. 1) having the voice concealment device 10 according to the first embodiment. The voice input unit 21 outputs the digital voice data generated from the input voice data to the echo canceller 22 in sampling units.

The echo canceller 22 has a function of suppressing and removing acoustic echo. The acoustic echo to be suppressed or removed in the voice concealment device 20 is input voice data corresponding to the disturbing sound from the voice concealment device 20 collected by the voice input unit 21. For example, the echo canceller 22 removes the signal corresponding to the disturbing sound from the digital voice data input from the voice input unit 21 in sampling units, and outputs the digital voice data from which the disturbing sound has been removed to the voice recognition unit 12. do. As a result, the voice recognition unit 12 can sequentially execute voice recognition processing on the digital voice data that does not include the disturbing sound, and output the output phonemes to the determination unit 13.

The storage unit 14 in the voice concealment device 20 has a concealment information list 142. Similar to the confidential information list 141, the confidential information list 142 stores confidential information (keywords such as personal information) that the user of the voice concealing device 20 desires to conceal as a confidential information phoneme series. Further, in the confidential information list 142, the matching degree determination value is stored in association with each confidential information phoneme sequence. The degree of matching determination value is a value indicating the ratio of phonemes that are in front of the combined output phoneme sequence among the phonemes constituting the secret information phoneme sequence. As will be described in detail later, the determination unit 13 in the voice concealment device 20 uses the concordance degree determination value in the concealment information determination process.

The voice data concealment unit 25 executes a voice concealment process for concealing voice (voice to be concealed) including contents that may correspond to confidential information. In the voice concealment device 20 according to the present embodiment, concealment of the concealment target voice indicates that a predetermined disturbing sound is output by superimposing it on the concealment target voice. When the voice concealment device 20 outputs a disturbing sound, the disturbing sound is input to the information terminal device arranged in the vicinity of the voice concealing device 20 together with the concealed target voice. This makes it difficult for the information terminal device to recognize the confidential voice, and prevents the confidential voice from leaking onto the network via the information terminal device.

The voice data concealment unit 25 includes a waveform output device (an example of an interference sound output unit) 251 capable of outputting an interference sound and a speaker (an example of an interference sound output unit) that outputs an interference sound corresponding to the waveform generated by the waveform output device. ) 252, the disturbing sound output control unit 250 that controls the output of the disturbing sound from the speaker 252 by controlling the generation of the disturbing sound waveform in the waveform output device 251 based on the confidential information determination result by the determination unit 13. have. As the disturbing sound generated by the waveform generated by the waveform output device 251, for example, a sound used as a self-regulating sound in television broadcasting or the like (so-called beep sound), a buzzer sound, or the like is assumed. The disturbing sound output control unit 250 determines whether or not to output the disturbing sound based on the confidential information determination result derived by the determination unit 13, and controls the generation of the disturbing voice waveform in the waveform output device 251. The waveform output device 251 generates a disturbing voice waveform when the disturbing sound output control unit 250 determines that the disturbing sound can be output, and transmits a signal including the generated disturbing voice waveform to the speaker 252. As a result, the disturbing voice is output from the speaker 252 based on the secret information determination result, and the secret target voice is concealed. Further, the voice waveform corresponding to the disturbing voice waveform generated by the waveform output device 251 is input to the echo canceller 22 in advance. Therefore, the echo canceller 22 can remove the digital signal corresponding to the disturbing sound included in the input voice data from the digital voice data.

(2-1) Voice Concealment Processing Next, with reference to FIG. 6, an example of the flow of voice concealment processing in the voice data concealment unit 25 will be described with reference to FIG. 7. FIG. 7 is a flowchart illustrating a flow of voice concealment processing. The voice concealment process is a process that makes it difficult to recognize the concealed voice by controlling the disturbing sound output by the disturbing sound output control unit 250. In the present embodiment, the voice concealment processing is executed based on the fact that the concealment information determination processing result is input from the determination unit 13 to the voice data concealment unit 25. That is, in the voice concealment processing, the input voice data is input to the voice input unit 21, the acoustic echo is removed by the echo canceller 22, and the voice is output to the voice recognition unit 12 in sampling units. Is executed repeatedly every time.

(Step S31)
In step S31, the disturbing sound output control unit 250 receives the confidential information determination result from the determination unit 13, and proceeds to the process of step S32.

(Step S32)
In step S32, the disturbing sound output control unit 250 confirms the confidential information determination result based on the confidential information determination signal received from the determination unit 13, and the output phoneme that the voice recognition unit 12 recognizes the digital voice data at the phoneme level is generated. Confidential information Determines whether or not there is a prefix match with the phoneme sequence. When the disturbing sound output control unit 250 receives the high-level confidential information determination signal, it determines that the output phoneme is forward-matched with the confidential information phoneme sequence, and proceeds to the process of step S33. The disturbing sound output control unit 250 recognizes that the digital voice data to be processed corresponds to the secret target voice by receiving the secret information determination signal having a high signal level. On the other hand, when the disturbing sound output control unit 250 receives the low-level confidential information determination signal, it determines that the output phoneme does not match the confidential information phoneme sequence in the forward direction, and proceeds to the process of step S34. The disturbing sound output control unit 250 recognizes that the digital voice data to be processed does not correspond to the secret target voice because the confidential information determination signal having a low signal level is received.

(Step S33)
In step S33, the disturbing sound output control unit 250 outputs the disturbing sound from the waveform output device 251 and ends the voice concealment processing. The disturbing sound output control unit 250 continues to output the disturbing sound in the waveform output device 251 until it is determined in step S33 that the output phoneme does not forwardly match the confidential information phoneme sequence.

(Step S34)
In step S34, the disturbing sound output control unit 250 determines whether or not the disturbing sound is being output by the waveform output device 251. When the disturbing sound output control unit 250 determines that the disturbing sound is being output, the process proceeds to step S35. On the other hand, when the disturbing sound output control unit 250 determines that the disturbing sound is not being output, the voice concealment processing is terminated.

(Step S35)
In step S35, the disturbing sound output control unit 250 stops the output of the disturbing sound in the waveform output device 251 and ends the voice concealment processing.

As described above, in the voice concealment device 20 according to the present embodiment, the disturbing sound output control unit 250 of the voice data concealment unit 25 may match the keyword such as personal information with the input voice data by the determination unit 13 (for example). , When it is determined that the output sound is in front of the confidential information sound element sequence), the waveform output device 251 (an example of the disturbing sound output unit) is used to generate the disturbing voice waveform, so that the speaker 252 (jamming sound output) is generated. A disturbing sound (an example of a predetermined disturbing sound) based on the disturbing voice waveform is output from the part). As a result, the voice concealment device 20 conceals the concealed target voice that may include confidential information (for example, a keyword including personal information) registered by the user, and the concealed target voice is via an information terminal device having a voice recognition function. It is possible to prevent the information from being leaked (transmitted to a voice recognition server or the like) on the network.

(2-2) Confidential Information Determination Process Next, the confidential information determination process in the voice concealment device 20 according to the second embodiment will be described. When the degree of matching between the confidential information phoneme sequence and the output phoneme reaches the matching degree determination value (more than a predetermined degree), the determination unit 13 of the voice concealment device 20 contains the concealment information in the input voice data. Judge that it may be included. Specifically, in the confidential information determination process, the determination unit 13 combines the number of phonemes corresponding to the ratio indicated by the matching degree determination value among the phonemes constituting the confidential information phoneme sequence (to a continuous input voice). It is determined whether or not there is a prefix match with the combination of the corresponding output phonemes). For example, when there are 10 phonemes constituting the confidential information phoneme sequence and the matching degree judgment value is 60%, the first 6 (= 10 × 60%) phonemes of the confidential information phoneme sequence match the combined output phonemes. At that time, that is, when the degree of matching between the confidential information phoneme sequence and the output phoneme reaches 60%, it is determined that the input voice data and the confidential information may match. The degree of matching determination value can be set by the user to any value. Further, the matching degree determination value can be appropriately changed, and if it is not set by the user, a default value (for example, 60%) may be set.

In the confidential information list 142 shown in FIG. 6, "60%" is registered as a match degree determination value in association with the confidential information n "090012344567" indicating a telephone number. In this case, the determination unit 13 has 23 phonemes (the fractions are truncated), which is 60% of the 39 phonemes (zerokyuuzerozeroichiniisanyongorokunana) constituting the secret information phoneme sequence corresponding to the secret information n "0901234567", as combined output phonemes. When the sequence matches, the confidential information determination signal (individual) corresponding to the confidential information n is set to a high level, and the high-level confidential information determination signal (comprehensive) is output to the disturbing sound output control unit 250. In this way, the determination unit 13 calculates the number of phonemes of the secret information phoneme series, which is a criterion for determining whether or not there is a prefix match between the combined output phoneme and the secret information phoneme series, based on the matching degree determination value.

For example, it is assumed that the input voice "telephone number is 090012345667" including the content that completely matches the confidential information n is input to the voice concealment device 20. In this case, when the determination unit 13 determines that the combined output phoneme sequence corresponding to the digital voice data including the content "090012" is forwardly matched with the confidential information phoneme sequence of the confidential information n, the confidential information corresponding to the confidential information n is concealed. It is determined that the degree of prefix matching between the information phoneme sequence and the combined output phoneme sequence has reached 60%. Along with this, the determination unit 13 determines that the input voice is the confidential voice, sets the confidential information determination signal (individual) corresponding to the confidential information n to a high level, and sets the high level confidential information. The determination signal (comprehensive) is output to the disturbing sound output control unit 250. When the interference sound output control unit 250 of the interference sound output control unit 250 receives a high-level confidential information determination signal, the interference sound output control unit 250 controls the waveform output device 251 so as to generate an interference sound waveform, and the interference waveform signal from the waveform output device 251. The speaker 252 receiving the input of the above outputs the disturbing sound (step S33).
As a result, the disturbing sound is output by superimposing on the secret target voice "34567" which is the latter half of the secret information n in the series of input voices. Therefore, the disturbing sound is input to the information terminal device arranged in the vicinity of the voice concealment device 20 together with at least a part of the concealment target voice. Therefore, it becomes difficult for the information terminal device to recognize the voice to be concealed, and it is prevented that the confidential information is leaked to the network via the information terminal device. Further, by setting the matching degree determination value to a predetermined value or more (for example, 50% or more), the frequency of determining that the input voice data that does not match the confidential information may match the confidential information is reduced. Therefore, it is possible to suppress the generation of unnecessary disturbing sounds.

(Modified example of the second embodiment)
In the confidential information determination process, the determination unit 13 in the voice concealment device 20 uses the keyword (text information) input by the user before conversion into the concealment information phoneme sequence and the concealment degree determination value in the concealment information in the input voice data. It may be determined whether or not there is a possibility that is included. In this case, the determination unit 13 conceals the characters constituting the keyword in the input voice data when the number of characters corresponding to the ratio indicated by the matching degree determination value matches the text information derived from the input voice. It may be determined that the information may be included. In this example, the voice recognition unit 12 converts the digital voice data recognized at the phoneme level into text information (phoneme conversion text), and outputs the phoneme conversion text to the determination unit 13. Further, the keywords before being converted into the confidential information phoneme series are stored in the confidential information list 142. As a result, the determination unit 13 can determine whether or not the degree of prefix matching between the keyword and the phoneme conversion text has reached the ratio set in the matching degree determination value. For example, when the degree of matching determination value is 30%, the determination unit 13 inputs when 3 characters out of the 10-character keywords are prefix-matched with the phoneme conversion text, that is, when the degree of prefix matching is 30%. It is determined that the voice data may contain confidential information.

Although the present invention has been described above with reference to each embodiment, the scope of the present invention is not limited to the illustrated and described exemplary embodiments, and the present invention has an effect equivalent to that of the object of the present invention. Also includes all embodiments. Furthermore, the scope of the present invention is not limited to the combination of the features of the invention defined by the claims, but may be defined by any desired combination of specific features among all disclosed features.

1 Voice processing device 3 Network 9 Voice recognition service system 10, 20 Voice concealment device 11 Voice input device 12 Voice recognition unit 13 Judgment unit 14 Storage unit 15, 25 Voice data concealment unit 21 Voice input unit 22 Echo canceller 120 Engine unit 121 Model 141 Confidential information list 150 Output control unit 151 Buffer 100 Voice recognition server 250 Interfering sound output control unit 251 Waveform output device 500 Server

Claims (7)

A storage unit that can store phoneme sequence information in which multiple keywords are converted into sound information,
A voice recognition unit that outputs voice data obtained by converting input voice from a voice source into a digital signal as output phoneme information, which is information recognized at the phoneme level by sequential voice recognition processing.
Whether or not there is a possibility that the input voice and the keyword may match based on the degree of matching between the output phoneme information output by the voice recognition unit and the stored phoneme information which is the phoneme sequence information stored in the storage unit. A judgment unit that can sequentially determine whether or not
A voice data concealment unit capable of concealing the content of voice data corresponding to the input voice determined by the determination unit to be likely to match the keyword.
Equipped with
In the determination unit, the input voice and the keyword match based on the degree of prefix matching between the output phoneme information output by the voice recognition unit and the stored phoneme information which is the phoneme sequence information stored in the storage unit. It is possible to sequentially determine whether or not there is a risk of
The voice data concealment unit can conceal the content of the voice data corresponding to the output phoneme information whose degree of prefix matching with the stored phoneme information is determined by the determination unit to be equal to or higher than a predetermined degree. A voice concealment device featuring. A storage unit that can store phoneme sequence information in which multiple keywords are converted into sound information,
A voice recognition unit that outputs voice data obtained by converting input voice from a voice source into a digital signal as output phoneme information, which is information recognized at the phoneme level by sequential voice recognition processing.
Whether or not there is a possibility that the input voice and the keyword may match based on the degree of matching between the output phoneme information output by the voice recognition unit and the stored phoneme information which is the phoneme sequence information stored in the storage unit. A judgment unit that can sequentially determine whether or not
A voice data concealment unit capable of concealing the content of voice data corresponding to the input voice determined by the determination unit to be likely to match the keyword.
Equipped with
The voice recognition unit discriminates between a voice section in which voice exists and a silent section in which voice does not exist in the voice data obtained by converting the input voice.
The voice data concealment unit is input from the voice source within a predetermined fixed period after the determination based on the determination by the determination unit that the input voice may match the keyword. Conceal the content of the voice data corresponding to the input voice, or the content of the voice data in a predetermined number of voice sections among the voice data corresponding to the input voice input from the voice source after the determination.
A voice concealment device characterized by this. The voice recognition unit discriminates between a voice section in which voice exists and a silent section in which voice does not exist in the voice data obtained by converting the input voice.
The voice data concealment unit is input from the voice source within a predetermined fixed period after the determination based on the determination by the determination unit that the input voice may match the keyword. It is characterized in that the content of the voice data corresponding to the input voice or the content of the voice data in a predetermined number of voice sections among the voice data corresponding to the input voice input from the voice source after the determination is concealed. The voice concealment device according to claim 1 . The voice data concealment unit is characterized in that the content of the voice data corresponding to the input voice is not concealed when the determination unit determines that the input voice is unlikely to completely match the keyword. The voice concealment device according to any one of claims 1 to 3 . The voice data concealment unit is
A buffer that can temporarily store the voice data converted from the input voice, and
It has a buffer output control unit that controls the output of audio data stored in the buffer based on the determination result by the determination unit.
One of claims 1 to 4, wherein the buffer output control unit does not output voice data corresponding to the input voice determined by the determination unit to be likely to match the keyword from the buffer. The voice concealment device according to item 1. The voice data concealment unit is
Interfering sound output unit that can output interfering sound,
It has an interfering sound output control unit that controls the output of the interfering sound from the interfering sound output unit based on the determination result by the determination unit.
The disturbing sound output control unit is characterized in that when it is determined by the determination unit that the keyword and the input voice may match, a predetermined disturbing sound is output to the disturbing sound output unit. The voice concealment device according to any one of claims 1 to 5. The specified information terminal device,
A storage unit that can store phoneme sequence information in which multiple keywords are converted into sound information.
A voice recognition unit that outputs voice data obtained by converting input voice from a voice source into a digital signal as output phoneme information, which is information recognized at the phoneme level by sequential voice recognition processing.
Whether or not there is a possibility that the input voice and the keyword may match based on the degree of matching between the output phoneme information output by the voice recognition unit and the stored phoneme information which is the phoneme sequence information stored in the storage unit. It functions as a determination unit that can sequentially determine whether or not, and a voice data concealment unit that can conceal the content of the voice data corresponding to the input voice that is determined by the determination unit to be likely to match the keyword .
In the determination unit, the input voice and the keyword match based on the degree of prefix matching between the output phoneme information output by the voice recognition unit and the stored phoneme information which is the phoneme sequence information stored in the storage unit. It is possible to sequentially determine whether or not there is a risk of
The voice data concealment unit can conceal the content of the voice data corresponding to the output phoneme information whose degree of prefix matching with the stored phoneme information is determined by the determination unit to be equal to or higher than a predetermined degree. A voice concealment program featuring.

Images