JP2000155600A - Speech recognition system and input voice level alarming method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To interactively given an alarm to a speaker so as to talke at a proper voice level when an input voice level is not within a proper range without necessitating additional circuits. SOLUTION: In a speech recognition system responding to a voice while recgnizing a speech, an input voice level judging part 10 judging levels of input voice data is provided so as to become the preceding stage of a speech recognizing part 20 in a CPU in which the part 20 is mounted and when the level of the input voice data is judged to be out of a reference range, this system is made to interactively perform the alarm of the result of the judgment and in the case the level is judged to be within the reference range and when the system is in the registration mode of the input voice data, the registring of the input voice data is performed by the recognizing part 20 and when the system is in the recognition mode of the input voice data, the recognizing of the speech is performed by the recognizing part 20.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice recognition system and an input voice level warning method, and more particularly to a voice recognition system for recognizing voice by inputting from a microphone and responding with voice from a speaker, and an input voice level warning in the voice recognition system. About the method.

[0002]

2. Description of the Related Art In a speech recognition system, it has been known that the magnitude of an input speech level (hereinafter, referred to as "high or low level") has a great effect on a recognition rate. That is, if the input audio level is too high, the input audio signal will be clipped and different from the original audio waveform. Conversely,
If the input audio level is too low, the input audio signal will be buried in ambient noise. As a result, the speech recognition rate decreases.

As a countermeasure for solving such a problem, a conventional voice recognition system includes an AGC (Automatic Control Unit) which detects an input voice level and controls the detected level to an appropriate range.
In some cases, an organic gain control circuit is provided.

FIG. 8 is a block diagram showing an example of a conventional speech recognition system. The voice recognition system includes a microphone 81 for converting voice into an analog voice signal, a microphone 8
Amplifier 8 for amplifying analog audio signal output from 1
2, an AGC circuit 83 that detects the level of the analog audio signal output from the amplifier 82 and controls the amplification factor of the amplifier 82, and an A / A converter that converts the analog audio signal output from the AGC circuit 83 into digital audio data. D converter 84
And a CPU (Central Proc) for inputting digital voice data from the A / D converter 84 and performing voice recognition.
An essential unit is constituted by an essing unit 85 and a memory 86 connected to the CPU 85.

FIG. 9 is a block diagram showing another example of a conventional speech recognition system. The voice recognition system includes a microphone 91 that converts voice into an analog voice signal,
An amplifier 92 for amplifying an analog audio signal output from the microphone 91, and an A / D converter 93 for converting the analog audio signal output from the amplifier 92 into digital audio data
And an AG for detecting the level of the digital audio signal output from the A / D converter 93 and controlling the amplification factor of the amplifier 92
A C circuit 94; a CPU 95 for inputting digital voice data from the AGC circuit 94 to perform voice recognition;
And a memory 96 connected to the main part.

[0006]

The first problem is that the conventional speech recognition system causes a reduction in the recognition rate. The reason is that if the amplification factor of the amplifier is automatically changed by the AGC circuit during the voice input, the waveform of the input voice signal is deformed, which adversely affects the recognition rate.

The second problem is that even if the amplification factor of the amplifier is controlled by the AGC circuit, if the amplification factor is not appropriate for the next speech input, speech recognition is not performed correctly without the speaker's knowledge. There is a risk. The reason is that if the gain of the amplifier is changed by the AGC circuit at the time of voice registration and at the time of recognition, the gain of the amplifier is different even if the speaker thinks that he / she spoke the same way. The reason is that there is a possibility that it is recognized as a different word.

[0008] The third problem is that it is costly. The reason is that when an AGC circuit is provided, the circuit is added accordingly.

An object of the present invention is to eliminate the need for an additional circuit,
It is an object of the present invention to provide a speech recognition system which, when an input speech level is not within an appropriate range, interactively gives a warning by voice so that a speaker speaks at an appropriate input speech level.

It is another object of the present invention to provide a voice recognition system in which a speaker is interactively warned by voice so as to speak at a proper input voice level when the input voice level is not within an appropriate range. To provide a method for warning an input voice level.

As a prior art document, Japanese Patent Laid-Open No. 2-1
No. 0500. The "voice recognition device" disclosed in this publication compares the input level with a threshold when a voice input is made, and when the input level is lower than the threshold, a short time means that the input voice is low. Warning sound is generated, and when the input level is higher than the threshold, a short-time warning sound is generated, which means that the input voice is louder, so that the voice itself gives an interactive warning. This is different from the present invention.

[0012]

A speech recognition system according to the present invention is a speech recognition system for recognizing speech and responding with speech so that a digital processing unit equipped with a speech recognition unit is located upstream of the speech recognition unit. An input audio level determining unit for determining the level of the input audio data, and when the input audio level determining unit determines that the level of the input audio data is out of the reference range, a warning to that effect is given by voice interactively. And when the input voice level determination unit determines that the level of the input voice data is within the reference range, the voice recognition unit performs voice recognition.

The speech recognition system of the present invention is a speech recognition system for recognizing speech and responding with speech. In a speech recognition system equipped with a speech recognition unit, the input speech data is provided so as to be in front of the speech recognition unit. An input audio level determining unit for determining the level of the input audio data is provided. When the input audio level determining unit determines that the level of the input audio data is out of the reference range, a warning to that effect is given interactively by voice. If the input voice level determination unit determines that the level of the input voice data is within the reference range, the input voice data is registered by the voice recognition unit if the input voice data is in the registration mode. In the voice data recognition mode, the voice recognition unit performs voice recognition.

Further, in the voice recognition system of the present invention, an analog voice signal input from a microphone is A / D converted by an A / D converter, and then input as input voice data to a digital processing unit for voice recognition. In a voice recognition system that responds with voice from a speaker, the digital arithmetic processing unit compares input voice data with upper-limit reference data and lower-limit reference data so that the level of the input voice data is higher than that of a voice recognition unit. Reference data comparing means for determining whether or not the input voice data is within the reference range; and, when the reference data comparing means determines that the level of the input voice data is not within the reference range, a sound is issued from the speaker to warn of the fact. An input voice level determination unit comprising a warning utterance means for causing the input voice level is provided.

Still further, the speech recognition system of the present invention performs A / D conversion of an analog speech signal input from a microphone by an A / D converter, and then inputs the signal as input speech data to a digital arithmetic processing unit to perform speech recognition. And a reference data comparing means for comparing the input voice data with the upper reference data and the lower reference data to determine whether the level of the input voice data is within a reference range. When the reference data comparing means determines that the level of the input voice data is not within the reference range, a warning voice generating means for uttering a voice warning of the fact from a speaker, and the reference data comparing means If it is determined that the level is within the reference range, the input audio data is subjected to fast Fourier transform in frame units and Frequency spectrum converting means for decomposing into frequency components to obtain a frequency spectrum, change amount calculating means for calculating a change amount for each frame with respect to the frequency component converted into a frequency spectrum by the frequency spectrum converting means, and input voice data registration mode Registration / recognition mode discriminating means for discriminating whether the mode is a recognition mode or not, and a frequency component decomposed by the frequency spectrum converting means when the registration / recognition mode discriminating means discriminates the registration mode and the change. A time direction compressing means for compressing the change amount for each frame calculated by the amount calculating means in the time direction and registering it as registered voice data; and a case where the registration / recognition mode determining means determines that the recognition mode is set. The input voice data and the registered voice data are collated in frame units by the DP matching method, and the distance is determined. Matching means for calculating a value, a matching result determining means for comparing the distance value calculated by the matching means with a pre-registered threshold value to determine whether or not the distance value is equal to or smaller than the threshold value; A response uttering means for uttering response voice data in response to the registered voice data as a recognition result from the speaker as the recognition result of the registered voice data whose distance value is determined to be equal to or less than the threshold value by the result determining means. .

On the other hand, according to the input audio level warning method of the present invention, an analog audio signal input from a microphone is A / D converted by an A / D converter, and then input to a digital arithmetic processing device as input audio data to perform voice recognition. In the input voice level warning method of the voice recognition system that responds by voice from the speaker, the input voice data is compared with the upper reference data and the lower reference data to determine whether the level of the input voice data is within the reference range. Determining a reference data comparing step; and, when the reference data comparing means determines that the level of the input audio data is not within the reference range, issuing a warning sound from a speaker to warn the user of this. Features.

Further, the input audio level warning method of the present invention compares the input audio data with the upper reference data and the lower reference data to determine whether the level of the input audio data is within a reference range. Steps and
A warning uttering step of, when it is determined by the reference data comparing step that the level of the input voice data is not within the reference range, uttering a voice to warn the user from the speaker, and a level of the input voice data by the reference data comparing step Is determined to be within the reference range, the input audio data is subjected to fast Fourier transform in frame units, decomposed into frequency components, and converted into a frequency spectrum. A change amount calculation step of calculating a change amount of each frequency component for each frame, a registration / recognition mode determination step of determining whether the input voice data is in a registration mode or a recognition mode, and a registration / recognition mode determination step. If it is determined that the mode is the registration mode, A time-direction compression step of compressing in the time direction the frequency component decomposed in the spectralization step and the change amount of each frame calculated in the change amount calculation step and registering the same as registered voice data; A matching step of calculating the distance value by comparing the input voice data and the registered voice data in a frame unit by the DP matching method when it is determined by the step that the recognition mode is set, and a distance value calculated by the matching step. A comparison result determination step of comparing the distance value with a pre-registered threshold value to determine whether the distance value is equal to or less than the threshold value; and a registration in which the distance value is determined to be less than or equal to the threshold value by the comparison result determination step. A response utterance that utters response voice data from a speaker in response to the registered voice data as voice data as a recognition result. Characterized in that it comprises a step.

[0018] On the other hand, the recording medium of the present invention allows the computer to compare the input audio data with the upper reference data and the lower reference data to determine whether or not the level of the input audio data is within a reference range. Means for outputting, when the reference data comparing means determines that the level of the input voice data is not within the reference range, a voice for issuing a warning to that effect from a speaker, and warning means for outputting the input voice data by the reference data comparing means. Frequency spectrum converting means for performing fast Fourier transform of the input voice data in frame units when it is determined that the level is within the reference range, decomposing the frequency components into frequency components, and converting the frequency spectrum into frequency spectrums; A change amount calculating means for calculating a change amount of each frequency component for each frame, and an input audio data. Registration / recognition mode discriminating means for discriminating whether the mode is a registration mode or a recognition mode of the data. If the registration / recognition mode discriminating means discriminates the registration mode, the frequency decomposed by the frequency spectrum converting means. The time direction compression means for compressing the component and the change amount for each frame calculated by the change amount calculation means in the time direction and registering it as registered voice data, and the registration / recognition mode discrimination means determines the recognition mode. Matching means for comparing the input voice data and the registered voice data on a frame basis by the DP matching method to calculate a distance value, and comparing the distance value calculated by the matching means with a pre-registered threshold value Means for judging whether or not the distance value is equal to or smaller than a threshold value, and the distance value is determined by the collation result judging means. The program for functioning as a response utterance means for uttering voice response data in response to the registered voice data registered voice data is determined to be less as a recognition result from the speaker to record.

Further, the recording medium of the present invention allows a computer to compare input audio data with upper-limit reference data and lower-limit reference data to determine whether or not the level of the input audio data is within a reference range. A step for issuing a warning sound from a speaker when the level of the input voice data is determined not to be within the reference range in the reference data comparing step; and a level for the input voice data in the reference data comparing step. If it is determined that is within the reference range, the input voice data is subjected to Fast Fourier Transform in frame units, decomposed into frequency components, and converted into a frequency spectrum. The amount of change to calculate the amount of change for each frame for the component Calculation step, a registration / recognition mode determination step for determining whether the input voice data is in a registration mode or a recognition mode, and a frequency spectrum conversion step when the registration / recognition mode determination step determines that the input voice data is in the registration mode. A time direction compression step of compressing in the time direction the frequency component decomposed by the above and the change amount of each frame calculated in the change amount calculation step and registering the same as registered voice data; If it is determined that the distance value is equal to the distance, the input voice data and the registered voice data are compared in a frame unit by the DP matching method to calculate a distance value, and the distance value calculated in the matching step is registered in advance. Compare the threshold value to determine whether the distance value is below the threshold. And executing a response utterance step of causing the speaker to utter response voice data responding to the registered voice data as a recognition result using the registered voice data whose distance value is determined to be equal to or smaller than the threshold value in the collation result determination step. Record a program for

[0020]

Embodiments of the present invention will be described below in detail with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of the speech recognition system according to the first embodiment of the present invention. The speech recognition system according to the present embodiment includes a CPU 1, a memory 3, a microphone 4, an amplifier 5, an A / D converter 6,
The D / A converter 7, the amplifier 8, and the speaker 9 constitute the main part.

The CPU 1 includes an input voice level determination unit 10
And a voice recognition unit 20 are provided. The CPU 1 is, for example, a V83 which is a 32-bit CPU having an external operating frequency of 33 MHz and an internal operating frequency of 100 MHz.
1 (μPD705101) (manufactured by NEC Corporation) can be used.

The input voice level judging section 10 comprises reference data comparing means 11 and warning utterance means 12.

The speech recognition unit 20 includes a frequency spectrum conversion unit 21, a change amount calculation unit 22, a registration / recognition mode determination unit 23, a time direction compression unit 24, and a verification unit 25.
And a collation result determination unit 26 and a response utterance unit 27.

The memory 3 has a reference data area 31, a warning voice data area 32, a registered voice data area 33,
A threshold area 34 and a response voice data area 35 are provided.

The reference data comparing means 11 is a means for comparing the input voice data with the reference data in the reference data area 31 to determine whether the input voice data is within the reference range. As the reference data, the reference data for detecting the upper limit level (hereinafter, referred to as upper limit reference data) is 7FFFFH (H at the end indicates that it is a hexadecimal number; the same applies hereinafter) or 8.
000H (that is, when the voltage exceeds 2 V peak-to-peak), and the reference data for detecting the lower limit level (hereinafter referred to as lower limit reference data) is 1666H or E999H (when the peak-to-peak is 350 mV or less, for example).
As 6-bit data, the reference data area 31 of the memory 3
Is stored in advance. The input audio data compares the reference data in a raw form. That is, during one sampling, when the voltage exceeds 2 V even for a moment (that is, when clipping is performed), and when the voltage does not exceed 350 mV, the reference data comparison unit 11 passes control to the warning utterance unit 12.

When the reference data comparing means 11 determines that the input voice data is not within the reference range between the upper limit reference data and the lower limit reference data, the warning utterance means 12 outputs
This is a means for outputting the corresponding warning voice data stored in advance in the warning voice data area 32 to the D / A converter 7. Warning voice data is "too loud" if it is larger than the upper reference data, but "too loud" if it is less than the lower reference data
Are stored in advance in the warning voice data area 32 of the memory 3.

When the reference data comparing means 11 determines that the input audio data is within the reference range between the upper reference data and the lower reference data, the frequency spectrum converting means 21 converts the input audio data into frames. This is a means for performing fast Fourier transform, decomposing the frequency components, and converting them into a frequency spectrum. The input audio data is quantized at 11 kHz, and is subjected to a fast Fourier transform using, for example, 16 ms data as a frame unit to be converted into a frequency spectrum.

The change amount calculating means 22 is a means for calculating a change amount (cepstrum) for each frame with respect to the frequency components frequency-converted by the frequency spectrum converting means 21.

The registration / recognition mode determining means 23 determines whether the input voice data is in the registration mode or the recognition mode based on the operation of a switch (not shown) or the like.

When the registration / recognition mode determining unit 23 determines that the input voice data is in the registration mode, the time direction compression unit 24 reduces the frequency component and the amount of change for each frame by half in the time direction. This is a means for compressing and registering in the registered voice data area 33 of the memory 3 as registered voice data (such as a speaker and a word).

When the registration / recognition mode determining means 23 determines that the input voice data is in the recognition mode, the collating means 25 compares the input voice data with the registered voice data area 3.
3 and DP (Dynamic Pro)
(Gramming) This is a means for matching on a frame basis by a matching method. In the DP matching method, matching (distance calculation) is performed by associating a feature amount pattern obtained by analyzing input voice data with a feature amount pattern of registered voice data while associating them with each other on a time axis. This is a method of selecting audio data.

The collation result determination means 26 compares the distance value obtained by the collation means 25 with a threshold value stored in advance in the threshold area 34 of the memory 3 to determine whether the distance value is equal to or less than the threshold value. It is a means to determine whether or not.

The response utterance means 27 includes the collation result determination means 2
If there is any registered voice data whose distance value is determined to be equal to or smaller than the threshold value according to 6, the recognition result is obtained, and response voice data responding to the registered voice data is selectively read out from the response voice data area 35 of the memory 3. This is a means for outputting to the D / A converter 7. Note that the response utterance means 27 has a learning function (a function of giving a more responsive response as one speaks), a clock function (a function of responding to a time inquiry).

Referring to FIG. 2, the processing of the voice recognition system according to the first embodiment includes an input voice data input step S101, a reference data comparison step S102,
Warning audio data output step S103, input audio data writing step S104, frequency spectrum conversion step S105, change amount calculation step S106, registration mode determination step S107, time direction compression / registration step S108, data collation step S109 When,
It comprises a collation result determination step S110 and a response voice data output step S111.

Referring to FIG. 3, reference data comparison step S102 and warning voice data output step S103
The more detailed processing of step S201 determines whether the data is larger than the upper limit reference data, and the step of outputting warning voice data S2
02 and whether it is smaller than the lower limit reference data or not Step S2
03 and a warning voice data output step S204.

FIGS. 4A to 4D are diagrams for explaining the sequential steps at the time of registering the input voice data by the voice recognition unit 20. FIG.

FIG. 5 is a diagram for explaining a DP matching method between the input voice data and the registered voice data by the matching means 25.

Next, the operation of the thus configured speech recognition system according to the first embodiment will be described with reference to FIGS.

(1) Operation when registering input voice data

When the speaker speaks while the registration mode is selected by operating a switch or the like (not shown), the microphone 4
Inputs this and converts it to an analog audio signal,
Amplifies the analog audio signal.

The A / D converter 6 converts the analog audio signal amplified by the amplifier 5 into, for example, an 11 kHz PCM (P
The data is sampled and digitized by an ulse code modulation (ulse code modulation) method, and is converted into 16-bit input audio data. The format of the input audio data is expressed, for example, as 7FFFFH for an input voltage of 2's complement or more than positive full scale, 8000H for an input voltage of less than negative full scale, and 0000H for no input. You. The input voltage range is, for example, 2 peak-to-peak.
V.

When the CPU 1 reads the input voice data (step S101), the reference data comparison means 11 stores the digital voice data input from the A / D converter 6 and the reference data area 31 of the memory 3. The upper limit reference data and the lower limit reference data are compared (step S102).
Thus, the corresponding warning voice data in the warning voice data area 32 of the memory 3 is output to the D / A converter 7 (step S103).

More specifically, the reference data comparing means 11 determines whether the input voice data is larger than the upper limit reference data stored in the reference data area 31 (step S).
201), if it is loud, the warning utterance means 12 reads out the corresponding warning voice data "the voice is too loud" from the warning voice data area 32 of the memory 3 and outputs it to the D / A converter 7 (step S201). S202).

Thus, the warning voice data is converted into an analog warning voice signal by the D / A converter 7, amplified by the amplifier 8, and emitted from the speaker 9 as a voice saying “the voice is too loud”. By listening to this warning sound, the speaker can know that the sound uttered by the user exceeds the proper input sound level. You can re-enter.

Next, the reference data comparing means 11 determines whether the input voice data is smaller than the lower limit reference data stored in the reference data area 31 (step S20).
3) If small, the warning uttering means 12
The corresponding warning voice data indicating that "the voice is too low" is read out from the warning voice data area 32 and output to the D / A converter 7 (step S204).

Thus, the warning voice data is converted into an analog warning voice signal by the D / A converter 7, amplified by the amplifier 8, and emitted from the speaker 9 as a voice saying “the voice is too low”. By listening to this warning voice, the speaker can know that the voice uttered by himself is lower than the proper input voice level, so re-inputting the voice at the proper input voice level by raising the voice can do.

On the other hand, when the reference data comparing means 11 determines that the input voice data is not larger than the upper limit reference data and not smaller than the lower limit reference data, that is, the input voice data is between the upper limit reference data and the lower limit reference data. Is determined to be within the reference range of
Writes 16-bit input audio data into a work area (not shown) of the memory 3 for each sampling (step S104).

Next, the CPU 1 causes the frequency spectrum converting means 21 to perform the following operations as shown in FIGS. 4 (a) and 4 (b).
The input voice data is subjected to fast Fourier transform, decomposed into frequency components, and converted into a frequency spectrum (step S105). For example, 16-bit input audio data is collected for 16 ms (called a frame), and all the input audio data is decomposed into frequency components.

Subsequently, the CPU 1 changes the change amount calculating means 22.
As shown in FIGS. 4B and 4C, the change amount (cepstrum) for each frame is calculated (step S).
106).

Next, the CPU 1 uses the registration / recognition mode determining means 23 to determine whether the input voice data is in the registration mode or the recognition mode (step S107).

Since the input voice data is in the registration mode, the CPU 1 causes the time direction compression means 24 to
As shown in (c) and (d), the frequency component and the amount of change for each frame are compressed to half in the time direction and registered in the registered audio data area 33 of the memory 3 as registered audio data (step S108). ). Thus, the registration of the input voice data (the registration of the speaker, the word, and the like) ends.

(2) Operation when collating input voice data

When the speaker speaks while the recognition mode is selected by operating a switch or the like (not shown), the microphone 4 inputs the same and converts it into an analog audio signal in the same manner as in the registration mode. 5 amplifies the analog audio signal, and the A / D converter 6 converts it into 16-bit input audio data.

When the input voice data is input (step S101), the reference data comparing means 11 compares the input voice data with the upper limit reference data and the lower limit reference data stored in the reference data area 31 of the memory 3. (Step S102).

The process of comparing the input voice data with the upper-limit reference data and the lower-limit reference data is exactly the same as in the case of the registration mode, and the input voice data is compared with the reference range between the upper-limit reference data and the lower-limit reference data. If it is determined that the voice is not within the range, the voice “loud voice” or “voice too low” is issued from the speaker 9 as a warning. Here, detailed description of the operation is omitted to avoid duplication.

When the reference data comparing means 11 determines that the input voice data is within the reference range between the upper limit reference data and the lower limit reference data, the CPU 1 sets the sampling rate for each sampling as in the case of the registration mode. Then, 16-bit input voice data is written in a work area (not shown) (step S104), and the input voice data is subjected to fast Fourier transform by the frequency spectrum converting means 21 to be decomposed into frequency components to generate a frequency spectrum (step S105).
The change amount calculating section 22 calculates the change amount (cepstrum) for each frame (step S106).

Next, the CPU 1 determines whether the mode is the registration mode or the recognition mode of the input voice data by the registration / recognition mode determining means 23 (step S107).

Since the input voice data is in the recognition mode, the CPU 1 uses the matching means 25 to input the input voice data in the registered voice data area 33 for each frame using the DP matching method as shown in FIG. It is checked whether the frame is close to the frame of the registered voice data, a difference from the closest frame is obtained, this difference is calculated for all frames of the input voice data, and the total of all the differences is set as a distance value (step S10).
9).

Next, the CPU 1 checks the collation result determining means 26.
Thus, if the distance value is larger than the threshold value stored in the threshold area 34 of the memory 3 in advance, it is determined that they do not match, and if the distance value is equal to or less than the threshold value, it is determined that they match (step S110).

If it is determined by the collation result determination means 26 that they match, the CPU 1 sets the response utterance means 27
Thereby, appropriate response voice data in the response voice data area 35 of the memory 3 is read and output to the D / A converter 7 (step S111).

As a result, the response voice data is converted into an analog voice signal by the D / A converter 7, amplified by the amplifier 7, and output as a response voice from the speaker 9. For example, in response to an input voice of "Hello", a response voice such as "How are you?" By listening to the response voice, the speaker can know that the voice recognition system is interactively responding to the voice uttered by the speaker.

If the collation result judging means 26 judges that there is no match, the CPU 1 means that the registered voice data (speaker, word, etc.) corresponding to the input voice data has not been registered. Ignore it without doing anything.

FIG. 6 shows the reference data comparison step S102 and the warning voice data output step S10 shown in FIG.
13 is a flowchart illustrating a modification of the third more detailed process. In the more detailed processing of the reference data comparison step S102 and the warning voice data output step S103 shown in FIG. 3, the input voice data is compared with the upper reference data and the lower reference data, and if the input voice data is not within the reference range, the warning voice data is output. However, in the more detailed processing of the reference data comparison step S102 and the warning voice data output step S103 shown in FIG. 6, the three reference values of the first reference data, the second reference data, and the third reference data are used. Data is provided, and the level of the input voice data is divided into four ranges to output different first warning voice data, second warning voice data, third warning voice data, and fourth warning voice data. . That is, the more detailed processing of the reference data comparison step S102 and the warning voice data output step S103 shown in FIG. 6 includes a first reference data comparison step S301, a first warning voice data output step S302, and a second reference data comparison step. S
303 and a second warning voice data output step S304
, A third reference data comparison step S305, a third warning voice data output step S306, and a fourth warning voice data output step S307.

Reference data comparison step S1 shown in FIG.
02 and the warning voice data output step S103 are described in detail in the reference data comparison step S103 shown in FIG.
In the case where the processing in step S103 and the warning voice data output step S103 are performed in more detail, the control is shifted to step S104 in FIG. 2 after the processing in FIG.
In such a case, the voice recognition by the voice recognition unit 20 is performed even if the level of the input voice data is out of the appropriate range.

For example, when the input audio data ranges from 0000H to FFFFH, the first reference data is set to 4000
H, the second reference data is 8000H, and the third reference data is C
000H and the level of the input audio data is 0000
H-3FFFFH, 4000H-7FFFH, 8000H
~ BFFFH, and C000H ~ FFFFH, the warning sound can be divided into four ranges.

Next, a second embodiment of the present invention will be described with reference to the drawings.

Referring to FIG. 7, the speech recognition system according to the second embodiment of the present invention is different from the speech recognition system according to the first embodiment shown in FIG. And a recording medium 40 on which a voice recognition program is recorded. This recording medium 40 may be a magnetic disk, a semiconductor memory, or another recording medium.

The input voice level determination program and the voice recognition program are read by the CPU 1 from the recording medium 40, and control the operation of the CPU 1 as the input voice level determination unit 10 and the voice recognition unit 20. The operations of the CPU 1 and the like under the control of the input voice level determination unit 10 and the voice recognition unit 20 are exactly the same as those in the case of the voice recognition system according to the first embodiment.

In each of the above embodiments, the microphone 4 and the amplifier 5 have been described as being different from each other.
Needless to say, the amplifier 5 and the amplifier 5 may be integrated.

Although the amplifier 8 and the speaker 9 have been described as being separate from each other, it goes without saying that the amplifier 8 and the speaker 9 may be integrated.

Further, amplifier 5, A / D converter 6, CP
Although the U1, D / A converter 7, and amplifier 8 have been described as separate units, these units are one LSI (Large S).
Needless to say, the information may be integrated as a call integrated circuit (call integrated circuit).

Further, an amplifier 5, an A / D converter 6,
CPU 1, memory 3, D / A converter 7, and amplifier 8
However, it is needless to say that these may be integrated as a so-called system LSI.

In the above embodiments, the case where the speech recognition system is a specific speaker recognition system has been described as an example. However, it goes without saying that the present invention can be similarly applied to a general-purpose speaker recognition system. Nor. That is, the feature of the present invention resides in that the input speech level determination unit is provided in the arithmetic processing device equipped with the speech recognition unit so as to be in front of the speech recognition unit. It does not matter whether it is something like that.

[0075]

The first effect is that the waveform of the input voice signal is not deformed during the voice input, so that the recognition rate does not decrease. The reason is that the AGC circuit is not used, and the level of the input voice signal does not change during the voice input, so that the recognition rate is not adversely affected.

The second effect is that voice input within the reference range can be easily performed. The reason is that by interactively warning the speaker that the input voice level is low or high by voice, it is possible to prompt the speaker to re-input voice within the reference range.

The third effect is that no additional circuit is required and the cost is not increased. The reason is that an arithmetic processing unit used for speech recognition in a general speech recognition system also detects an input speech level and performs a warning.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a speech recognition system according to a first embodiment of the present invention.

FIG. 2 is a flowchart showing processing of the speech recognition system according to the first embodiment.

FIG. 3 is a flowchart showing more detailed processing of a reference data comparison step and a warning voice data output step in FIG. 2;

4 (a) to 4 (d) are diagrams for explaining sequential steps when registering input voice data by a voice recognition unit in FIG. 1;

FIG. 5 is a diagram for explaining matching by the DP matching method performed by the matching means in FIG. 1;

FIG. 6 is a flowchart showing a modification of the more detailed processing of the reference data comparison step and the warning voice data output step shown in FIG. 2;

FIG. 7 is a block diagram illustrating a configuration of a speech recognition system according to a second embodiment of the present invention.

FIG. 8 is a block diagram showing an example of a conventional speech recognition system.

FIG. 9 is a block diagram showing another example of a conventional speech recognition system.

[Explanation of symbols]

 Reference Signs List 1 CPU 3 memory 4 microphone 5 amplifier 6 A / D converter 7 D / A converter 8 amplifier 9 speaker 10 input audio level determination unit 11 reference data comparison unit 12 warning utterance unit 20 voice recognition unit 21 frequency spectrum conversion unit 22 change Quantity calculation means 23 Registration / recognition mode discrimination means 24 Time direction compression means 25 Verification means 26 Verification result determination means 27 Response utterance means 31 Reference data area 32 Warning voice data area 33 Registered voice data area 34 Threshold area 35 Response voice data Area 40 Recording medium S101 Input voice data input step S102 Reference data comparison step S103 Warning voice data output step S104 Input voice data writing step S105 Frequency spectrum conversion step S106 Change amount calculation step S107 Registration mode discrimination Step S108 Time direction compression / registration step S109 Data collation step S110 Collation result determination step S111 Response voice data output step S201 Determination step larger than upper limit reference data S202 Warning voice data output step S203 Determination step lower than lower limit reference data S204 Warning voice Data output step S301 First reference data comparison step S302 First warning voice data output step S303 Second reference data comparison step S304 Second warning voice data output step S305 Third reference data comparison step S306 Third warning voice data output step S307 4 Warning audio data output step

Claims (8)

[Claims] 1. A speech recognition system for recognizing speech and responding with speech, wherein an input speech level for determining a level of input speech data in a digital processing unit equipped with a speech recognition unit so as to be a stage preceding the speech recognition unit. Provide a judgment unit,
When the level of the input voice data is determined to be out of the reference range by the input voice level determination unit, a warning to that effect is made interactively by voice, and the level of the input voice data is determined by the input voice level determination unit. A voice recognition system for performing voice recognition by the voice recognition unit when it is determined that is within a reference range. 2. A speech recognition system for recognizing speech and responding with speech, wherein an input speech level for determining a level of input speech data in a digital processing unit equipped with a speech recognition unit so as to be a stage preceding the speech recognition unit. Provide a judgment unit,
When the level of the input voice data is determined to be out of the reference range by the input voice level determination unit, a warning to that effect is made interactively by voice, and the level of the input voice data is determined by the input voice level determination unit. Is determined to be within the reference range, if the input voice data is in the registration mode, the input voice data is registered by the voice recognition unit.If the input voice data is in the recognition mode, the input voice data is registered by the voice recognition unit. A voice recognition system characterized by performing voice recognition. 3. An analog audio signal input from a microphone is A / D-converted by an A / D converter, and then input as input audio data to a digital processing unit to perform voice recognition.
In a voice recognition system that responds by voice from a speaker, the digital arithmetic processing device compares input voice data with upper-limit reference data and lower-limit reference data so that the level of the input voice data is higher than that of a voice recognition unit. A reference data comparing means for determining whether the input voice data is within the reference range; and, when the reference data comparing means determines that the level of the input voice data is not within the reference range, a sound is issued from the speaker to warn the user. A speech recognition system comprising an input speech level determination unit comprising warning utterance means. 4. An analog audio signal input from a microphone is A / D-converted by an A / D converter, and then input as input audio data to a digital processing unit to perform voice recognition.
In a voice recognition system that responds by voice from a speaker, reference data comparing means for comparing input voice data with upper limit reference data and lower limit reference data to determine whether or not the level of the input voice data is within a reference range. When the reference data comparing means determines that the level of the input voice data is not within the reference range, a warning voice generating means for uttering a voice to warn the user of the fact from the speaker, and the level of the input voice data is reduced by the reference data comparing means. Frequency spectrum converting means for performing fast Fourier transform on a frame basis to decompose the input voice data into frequency components when it is determined that the input voice data is within the reference range, and converting the frequency spectrum into a frequency spectrum by the frequency spectrum converting means; A change amount calculating means for calculating a change amount of each component for each frame, and an input Registration / recognition mode determining means for determining whether the mode is a voice data registration mode or a recognition mode, and when the registration / recognition mode determining means determines that the mode is the registration mode, the voice / speech data is decomposed by the frequency spectrum converting means. A time direction compression unit for compressing the frequency component and the change amount for each frame calculated by the change amount calculation unit in the time direction and registering them as registered voice data; and a recognition mode by the registration / recognition mode discrimination unit. When it is determined that the input voice data and the registered voice data are matched by a frame unit using the DP matching method to calculate a distance value, a threshold value calculated by the matching means and a threshold value registered in advance are used. Comparison result determining means for comparing the distance value with the value to determine whether the distance value is equal to or less than a threshold value;
Response speech means for uttering response speech data from the speaker as response to the registered speech data for which the distance value is determined to be less than or equal to the threshold value by the comparison result decision means as a recognition result. Speech recognition system. 5. An analog audio signal input from a microphone is A / D-converted by an A / D converter, and then input as input audio data to a digital processing unit to perform voice recognition.
In an input voice level warning method of a voice recognition system that responds by voice from a speaker, a criterion for comparing input voice data with upper reference data and lower reference data to determine whether the level of the input voice data is within a reference range. A data comparing step; and a warning uttering step of, when the reference data comparing means determines that the level of the input voice data is not within the reference range, uttering a voice to warn of the fact from a speaker. Input audio level warning method. 6. A reference data comparing step of comparing input voice data with upper limit reference data and lower limit reference data to determine whether or not the level of the input voice data is within a reference range. When it is determined that the level of the audio data is not within the reference range, a warning utterance step of uttering a voice to warn the user from the speaker, and when the level of the input audio data is within the reference range by the reference data comparison step. If determined, the input voice data is subjected to Fast Fourier Transform on a frame-by-frame basis, decomposed into frequency components and converted into a frequency spectrum, and the frequency component converted into a frequency spectrum by the frequency spectrum conversion step is changed for each frame. A change amount calculating step for calculating the amount, A registration / recognition mode determining step of determining whether the mode is the registration mode or the recognition mode, and a frequency component decomposed by the frequency spectrum conversion step when the registration / recognition mode determination step determines that the mode is the registration mode. And a change amount for each frame calculated in the change amount calculating step in the time direction, and registering as registered voice data in a time direction, and the registration / recognition mode determining step determines that the recognition mode is set. In this case, the input voice data and the registered voice data are compared in a frame unit by the DP matching method to calculate a distance value, and the distance value calculated in the matching step and a pre-registered threshold value are compared with each other. A comparison result determination step of comparing the distance value to determine whether the distance value is equal to or smaller than a threshold value; A response uttering step of uttering from a speaker response voice data responding to the registered voice data as a recognition result of the registered voice data whose distance value is determined to be equal to or smaller than the threshold value in the determining step. Audio level warning method. 7. A reference data comparing means for comparing the input audio data with the upper reference data and the lower reference data to determine whether or not the level of the input audio data is within a reference range. When the level of the input voice data is determined to be out of the reference range, a warning uttering means for uttering voice to warn the user from the speaker, and the level of the input voice data is within the reference range by the reference data comparing means. If it is determined that the input speech data is fast Fourier-transformed in frame units, decomposed into frequency components and converted into a frequency spectrum, frequency-spectrum-converted frequency components by this frequency-spectralizing means are changed for each frame Change amount calculation means for calculating the amount of input voice data / To determine if the mode is
Recognition mode discriminating means, when the registration / recognition mode discriminating means discriminates the registration mode, the frequency component decomposed by the frequency spectrum converting means and the change amount for each frame calculated by the change amount calculating means. Time-direction compression means for compressing the input speech data in the time direction and registering it as registered speech data; Matching means for calculating a distance value by matching the unit, comparing the distance value calculated by the matching means with a pre-registered threshold value to determine whether the distance value is equal to or less than the threshold value Means and the registered voice data whose distance value is determined to be equal to or less than the threshold value by the matching result determining means as the recognition result. Recording medium for recording a program for functioning as a response utterance means for utterance from the speaker response voice data responsive to audio data. 8. A reference data comparing step for comparing the input voice data with the upper reference data and the lower reference data to determine whether or not the level of the input voice data is within a reference range. When the level of the input voice data is determined not to be within the reference range, a warning utterance step of uttering voice to warn the user from the speaker, and the level of the input voice data is within the reference range by the reference data comparison step. If it is determined that the input voice data is fast Fourier-transformed in frame units and decomposed into frequency components to generate a frequency spectrum, the frequency component converted into a frequency spectrum by the frequency spectrum conversion step is changed for each frame. Change calculation step to calculate the amount, input voice A registration / recognition mode determination step for determining whether the mode is a data registration mode or a recognition mode, and a frequency decomposed by the frequency spectrum conversion step when the registration / recognition mode determination step determines that the mode is the registration mode. The recognition mode is determined by the registration / recognition mode determination step in the time direction compression step in which the component and the change amount for each frame calculated in the change amount calculation step are compressed in the time direction and registered as registered voice data. A matching step of calculating the distance value by comparing the input voice data with the registered voice data in a frame unit by the DP matching method, and comparing the distance value calculated in the matching step with a previously registered threshold value Matching result determination step for determining whether the distance value is equal to or smaller than a threshold value, and A program for executing a response utterance step of causing a speaker to utter response voice data responding to the registered voice data as a recognition result with respect to the registered voice data whose distance value is determined to be equal to or smaller than the threshold value in the comparison result determination step. The recording medium on which it was recorded.