JP2006154531A - Audio speed conversion device, audio speed conversion method, and audio speed conversion program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a device, a method, and a program for speech speed conversion that can accurately calculate the speech speed of a speaker's speech included in a speech signal to convert the speech speed of a reproduced speech that a user desires. <P>SOLUTION: A pause detection portion 12 decides a speech section and a non-speech section (pause) from a speech signal stored in a speech signal storage portion 11. A statistical data storage portion 13 stores statistical data regarding predetermined pause lengths. A speech speed calculation portion 14 measures pause lengths of respective pauses detected by the pause detection portion 12 and calculates the speech speed based upon the pause lengths and the statistical data stored in the statistical data storage portion 13. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an audio speed conversion apparatus, an audio speed conversion method, and an audio speed conversion program. More specifically, the present invention relates to an audio speed conversion apparatus, an audio speed conversion method, and an audio capable of converting a speech speed of an audio signal. It relates to a speed conversion program.

  2. Description of the Related Art Conventionally, a method is known in which a speech rate is converted with a constant compression / expansion rate at a rate at which a speaker speaks (hereinafter referred to as a speech rate). For example, when reproducing a speech signal of a conversation, the speech speed of an actual speaker can be changed at a constant compression / expansion rate to change the speech speed of the reproduced speech included in the speech signal. Here, in an actual conversation, a plurality of speakers may speak at different speaking speeds. In addition, even the same speaker may speak at different speaking speeds. In other words, the actual speaking speed of the speaker is often not constant. Therefore, when the speaking speed of the actual speaker is different, the speaking speed of the reproduced voice that has been converted at a certain compression / expansion rate may be faster or slower than the speaking speed desired by the user. As a result, there is a problem in that there are portions where it is difficult for the user to hear the voice of the speaker.

  Therefore, a method has been proposed in which the speech speed of an actual speaker is detected and speech speed conversion is performed by setting a compression / decompression rate corresponding to the speech speed (see, for example, Patent Document 1). The audio compression / decompression apparatus disclosed in Patent Document 1 will be described below with reference to FIG.

  In Patent Document 1, the actual speaking speed of a speaker is defined by the number of syllables per unit time and is referred to as “speech rate”. Here, a syllable is a group of phonemes (eg, vowels) with a certain voice length, or a group that can follow a very short phoneme (eg, consonants) before and / or after a phoneme with a certain voice length. means.

  FIG. 17 is a block diagram showing a configuration in which an audio compression / decompression apparatus is applied to an IC recorder. 17, the IC recorder 200 includes a microphone 207, an A / D converter 208, an IC memory 201, a compression / decompression device 206, a D / A converter 205, and a speaker 209.

  The IC recorder 200 converts an analog audio signal input from the microphone 207 into a digital signal in the A / D converter 208 when recording the voice of the speaker, and records the converted digital audio signal in the IC memory 201. . Also, the IC recorder 200 compresses and expands the digital signal of the voice recorded in the IC memory 201 on the time axis in the compression / decompression device 206 when reproducing the voice of the speaker. Thereafter, the compressed / expanded audio digital signal is converted into an analog signal by the D / A converter 205, and the compressed / expanded audio analog signal is reproduced from the speaker 209.

  The compression / decompression device 206 includes an utterance speed detection unit 202, a compression / expansion rate adjustment unit 203, and a pitch expansion / compression unit 204. Typically, the compression / decompression device 206 is configured by a DSP. The utterance speed detection unit 202 generates a time axis waveform of the audio signal from the audio digital signal recorded in the IC memory 201, and performs a smoothing process on the envelope of the time axis waveform. And the peak position of the waveform which comprises each syllable is detected for every predetermined time from the waveform by which the smoothing process was carried out, and the number of peaks is measured. Thereafter, the number of syllables per unit time obtained by dividing the number of peaks by the number of syllables and dividing the number of syllables by a predetermined time length is calculated as the utterance speed. Here, the peak means a portion having the maximum level in the waveform constituting each syllable.

  The compression / decompression rate adjustment unit 203 adjusts the compression / decompression rate based on the speech rate calculated by the speech rate detection unit 202. As an example, when the utterance speed is 8 syllables / second, the reference value for adjustment is used. When the utterance speed is the above-mentioned reference value, the compression / decompression rate is 50% for double-speed playback and 50% for 0.5-speed playback. When the utterance speed is greater than the reference value, the compression / decompression rate is set to less than 50% for double speed playback and to 50% or more for 0.5 speed playback. That is, the reproduction is adjusted so as to be delayed with respect to the reproduction when the utterance speed is the reference value. When the utterance speed is smaller than the reference value and slow, the compression / decompression ratio is set to 50% or more for the double speed reproduction, and the insertion ratio is set to less than 50% for the 0.5 speed reproduction. That is, the playback speed is adjusted to be faster than the playback speed when the utterance speed is the reference value.

  The pitch expansion / compression unit 204 compresses / decompresses the audio digital signal recorded in the IC memory 201 on the time axis based on the information of the compression / expansion rate adjusted by the compression / expansion rate adjustment unit 203, thereby The voice speed of the person's voice is converted.

As described above, the voice compression / decompression apparatus 206 calculates the utterance speed based on the number of peaks of the time axis waveform constituting the voice of the speaker every predetermined time. Then, the speech rate conversion can be performed by adjusting the compression / decompression rate according to the calculated speech rate.
Japanese Patent Laid-Open No. 7-64597

  Here, an audio signal (for example, a television program, a radio program, or a recording medium) in which a signal other than the audio (for example, a music signal, a noise signal, etc.) is superimposed on the voice of the speaker to be subjected to speech speed conversion is recorded. When a voice signal of a movie or the like is input, the time axis waveform of the voice of the speaker included in the voice signal is a waveform on which signals other than the voice of the speaker are superimposed. For this reason, the peak position of the waveform constituting the syllable does not necessarily correspond to the peak position of the syllable in the actual speaker's voice in the voice signal in which a signal other than the voice is superimposed on the voice of the target speaker. There is a case. In other words, the peak position of the waveform constituting the syllable does not necessarily correspond to the syllable by human voice. However, the conventional speech compression / decompression device 206 detects the peak position of the waveform constituting the syllable, measures the number of peaks, and calculates the utterance speed based on the number of peaks. It is difficult to accurately calculate the speaking rate of the speaker's voice. As a result, there has been a problem that it is not possible to convert the voice speed of the reproduced voice desired by the user.

  Therefore, an object of the present invention is to accurately calculate the speech speed of the speaker's voice even if the voice signal is a signal other than the voice superimposed on the voice of the speaker subject to speech speed conversion. Another object of the present invention is to provide an audio speed conversion apparatus, an audio speed conversion method, and an audio speed conversion program capable of converting the speech speed of reproduced audio desired by a user.

  A first aspect of the present invention is an audio speed conversion apparatus for converting an audio signal including a voice of a speaker to be converted into an audio speed and reproducing the audio signal, and an audio section including the voice of the speaker from the audio signal And a non-speech segment that does not include the voice of the speaker, and a non-speech segment detector that detects the non-speech segment and a time length for each non-speech segment detected from the speech signal by the non-speech segment detector A non-speech segment length measurement unit that measures the speech rate, a speech rate calculation unit that computes the speech rate of the speaker in the speech signal based on the time length of each non-speech segment measured by the non-speech segment length measurement unit, A speech speed converting / reproducing unit for converting the speech signal to be reproduced according to the speech rate calculated by the speed calculating unit;

  A second aspect of the invention is an audio speed conversion device according to the first aspect of the invention, wherein the speech speed calculation unit calculates the occurrence frequency of the time length for the non-speech section measured by the non-speech section length measurement unit. The speech rate calculation unit includes a calculation unit, based on a statistical relational expression between the speech rate obtained in advance and the time length of the non-speech interval, and the speaker's speed corresponding to the maximum frequency in the occurrence frequency Calculate the speaking rate.

  A third aspect of the present invention is an audio speed conversion device subordinate to the second aspect of the present invention, which stores statistical data for time length statistical data of a non-voice interval that is statistically determined for each of a plurality of preset segments. The speech rate calculating unit further includes a plurality of non-speech segments according to the time length of the non-speech segment measured by the non-speech segment length measurement unit based on the statistical data stored in the statistical data storage unit. And a non-speech segment classification unit that selects one segment from the plurality of segments based on a predetermined condition, and the occurrence frequency calculation unit is a non-speech segment belonging to the category selected by the non-speech segment classification unit The occurrence frequency is calculated using the time length of the speech section.

  A fourth aspect of the invention is an audio speed conversion device subordinate to the third aspect of the invention, wherein the statistical data storage unit sets a plurality of sections according to the speech speeds classified into a plurality, and statistical data for each of the sections. And the non-speech segment classification unit is configured to perform the non-speech segment according to the time length of the non-speech segment measured by the non-speech segment length measurement unit based on the statistical data for each category in which the speech speed is classified. Are classified for each category, and the category in which the most non-voice segments are classified is selected.

  A fifth aspect of the invention is an audio speed conversion device according to the third aspect of the invention, in which the statistical data storage unit statistical data for the first section obtained by statistically obtaining the time length of the non-speech interval that occurs immediately after the reading. And statistical data for the second segment obtained by statistically calculating the time length of the non-speech interval that occurs immediately after the punctuation point, and the non-speech segment classification unit is configured for each of the first segment and the second segment. Based on the statistical data, according to the time length of the non-speech segment measured by the non-speech segment length measurement unit, classify the non-speech segment for each segment and select the first segment, and the occurrence frequency calculation unit, The occurrence frequency is calculated using the time length of the non-speech section belonging to the first category.

  A sixth aspect of the invention is an audio speed conversion device according to the third aspect of the invention, in which the statistical data storage unit includes a plurality of categorized utterance speeds, a first categorization that occurs immediately after reading, and a punctuation immediately after a phrase. A plurality of sections are set according to the combination with the generated second section, and statistical data of the length of the non-speech section statistically obtained for each section is stored, and the non-speech section classification unit includes a plurality of sections Based on the statistical data for each category, the non-speech segment is classified for each category according to the time length of the non-speech segment measured by the non-speech segment length measurement unit, and a category corresponding to the speech rate classified into a plurality of categories Of the combinations with the first segment, the segment that classifies the most non-voice segments is extracted to determine the segment for the speech rate, and the segment for the determined speech rate and the first segment are combined. Select a category.

  A seventh aspect of the invention is an audio speed conversion device subordinate to the third aspect of the invention, wherein the statistical data storage unit sets a plurality of sections in advance according to the characteristics of the speaker, and the statistics for each of the plurality of sections. The statistical data of the time length of the non-speech interval obtained automatically is stored.

  An eighth invention is an audio speed conversion device according to the first invention, wherein the speech rate calculation unit is detected by the non-speech segment detection unit using the time length measured by the non-speech segment length measurement unit. The non-speech segment is classified into a plurality of groups, and a non-speech segment classification unit that selects one from the plurality of groups, and the time length of the non-speech segment belonging to the group selected by the non-speech segment classification unit are generated. The utterance of the speaker according to the time length that is the maximum frequency in the occurrence frequency based on the occurrence frequency calculation unit that calculates the frequency and the statistical relational expression between the utterance speed and the time length of the non-speech interval obtained in advance An utterance speed conversion unit for calculating the speed.

  A ninth invention is an audio speed conversion device subordinate to the first invention, wherein the display unit and the speech speed conversion / playback unit calculate a playback time for converting the voice signal to be played back, and the playback time is calculated. And a reproduction time calculation unit that displays information indicating the above on the display unit.

  A tenth aspect of the present invention is an audio speed conversion device subordinate to the first aspect of the present invention, wherein the display unit and the speech speed conversion / reproduction unit calculate the reproduction speed at which the speech signal is converted by the speech speed and reproduce the same. And a reproduction speed calculation unit that displays information indicating the above on the display unit.

  An eleventh aspect of the present invention is an audio speed conversion method for reproducing an audio signal including a voice of a speaker to be subjected to an audio speed conversion by reproducing the audio speed, wherein the audio section includes the voice of the speaker from the audio signal. And a non-speech segment detected from the speech signal for a predetermined time by a non-speech segment detection step for detecting the non-speech segment and the non-speech segment detection step A non-speech interval length measurement step that measures the time length for each, and a speech rate calculation that calculates the speaking rate of the speaker in the speech signal based on the time length of each non-speech interval measured in the non-speech interval length measurement step And a speech speed converting / reproducing step of reproducing the speech signal by converting the speech speed according to the speech speed calculated by the speech speed calculating step.

  A twelfth aspect of the invention is an audio speed conversion program executed by a computer of an audio speed conversion apparatus that converts an audio signal including a voice of a speaker to be converted into an audio speed and reproduces the audio signal. A non-speech section detecting step for distinguishing between a speech section in which a speaker's voice is included from a speech signal and a non-speech section in which the speaker's speech is not included, and detecting the non-speech section; and a non-speech section detecting step Based on the time length of each non-speech interval measured in the non-speech segment length measurement step and the non-speech segment length measurement step for measuring the length of time for each non-speech segment detected from the speech signal for a predetermined time. An utterance speed calculation step for calculating the utterance speed of the speaker in the signal, and an utterance speed conversion for reproducing the speech signal according to the utterance speed calculated by the utterance speed calculation step. To execute and raw step.

  According to the first aspect of the invention, since the speaking rate of the speaker in the speech signal is calculated based on the time length of the non-speech interval included in the speech signal, the speech is converted into the speech of the speaker that is subject to speech speed conversion. The speech rate of the speaker's speech can be accurately calculated even if the speech signal is a signal superimposed with other signals.

  According to the second aspect, the occurrence frequency of the time length of each non-speech interval detected from the audio signal is calculated, and the maximum frequency of the occurrence frequency is obtained based on a statistical relational expression obtained in advance. Since the speech rate is calculated according to the time length, it is possible to calculate an accurate speech rate with respect to the time length of the non-speech section having variations.

  According to the third invention, irregular time length data can be removed by classifying the time length of each non-speech segment detected from the speech signal into a preset category, and more accurate speech The speed can be calculated.

  According to the fourth aspect, each non-speech segment detected from the speech signal is classified into speech speed categories (for example, three categories set as fast, normal, and slow). By selecting the most segment, it is possible to select a non-voice segment suitable for calculating an accurate speech rate.

  According to the fifth aspect, since the time length of the non-speech section that occurs immediately after the reading and the utterance speed are highly correlated, only the time length of the non-speech section that belongs to the first section indicating immediately after the reading is obtained. By using this, it is possible to calculate an accurate speech rate with higher accuracy.

  According to the sixth aspect of the invention, using the characteristic of the time length of the non-speech section that occurs immediately after a highly-correlated phrase in order to roughly distinguish the speech rate, it belongs to the second category indicating immediately after the phrase. The number is used to classify the speech rate into categories, and then the non-speech interval time length that occurs immediately after the reading and the speech rate belong to the first category indicating the immediately after the reading using the characteristic that is highly correlated. By using only the time length of the non-speech section, it is possible to calculate the speech rate more accurately and accurately.

  According to the seventh aspect, each non-speech segment detected from the speech signal is classified into a plurality of sections corresponding to the characteristics of the speaker, and according to the characteristics of the speaker to be subjected to the speech speed conversion. An optimal speech rate can be calculated.

  According to the eighth aspect of the invention, a plurality of groups are set, and the time length of each non-speech interval detected from the speech signal is classified into the group based on statistical data statistically obtained for each group. Then, by selecting a group having the largest number of classified non-speech segments, it is possible to select an appropriate non-speech segment in order to calculate the speech rate.

  According to the ninth aspect of the present invention, the reproduction time of the audio signal can be grasped in advance by inputting the utterance speed of the reproduction audio desired by the user.

  According to the tenth aspect of the present invention, the speech rate of the reproduced speech after the speech rate conversion can be grasped in advance by inputting the playback time of the speech signal desired by the user.

  Further, according to the audio speed conversion method and the audio speed conversion program of the present invention, the same effects as those of the above-described audio speed conversion device can be obtained.

  Before describing the embodiment of the present invention, the concept of the present invention will be described with reference to FIGS. In addition, the speed (speaking speed) at which the voice is uttered is defined by the number of mora per unit time or the inverse thereof in the present invention, and is referred to as “speech speed”. Here, the mora corresponds to a kana character unit. Further, a section including the voice of the speaker to be subject to speech speed conversion is defined as a voice section, and a section not including the speaker's voice is defined as a non-voice section. The non-voice segment is called “pause”.

  In general, the relationship described below is known as the relationship between pause and speech rate. First, there is a relationship that the total number of pauses in the same sentence decreases as the utterance speed increases. Secondly, there is a relationship in which the total length of pauses (hereinafter referred to as pause length) in the same sentence becomes shorter as the speech rate is higher. Third, at the same speech rate, there is a relationship that the pause length varies depending on the phrase attribute (for example, punctuation in a sentence) of the voice near the pause.

  In addition, Masaki and two others, 2002 Acoustical Society of Japan Spring Research Presentation, “Speech duration between different speech speeds and utterance styles in narrative reading” About Control ", Acoustical Society of Japan Proceedings, Acoustical Society of Japan, March 2002, 2-10-17, p. 297-298 (hereinafter referred to as Document 1). Hereinafter, the outline described in Document 1 will be described.

  In the above-mentioned document 1, story readings are recorded in different utterance speed categories (fast, normal, and slow), and the occurrence frequency of poses by pose length according to the utterance speed category is analyzed. FIG. 1 is a diagram showing an example of the occurrence frequency for pauses according to pause lengths according to speech speed classification (fast, normal, slow). In FIG. 1, the vertical axis indicates the frequency of pose occurrence in degrees, and the horizontal axis indicates the pose length in milliseconds. In FIG. 1, the pause length at which the occurrence frequency becomes the maximum value at each utterance speed tends to be shorter as the utterance speed is higher.

  Here, focusing on the phrase “.” And the reading “,” in the sentence, which are the phrase attributes of the speech in the vicinity of the pose, the pose is classified into the pose of the section immediately after the punctuation and the pose of the section immediately after the reading. FIG. 2 is a schematic diagram illustrating an example of a pause in a time axis waveform of a voice. As shown in FIG. 2, the phrase attribute of a phrase is classified as “between sentences”, and the pose immediately after the phrase is set as “between sentences”. In addition, the word attribute of the punctuation mark is classified as “within sentence”, and the pose immediately after the punctuation mark is set as “within sentence sentence”. Then, the tendency of each pose length of the intra-sentence pose and the inter-sentence pose according to the utterance speed category is analyzed.

  First, the tendency of the pose length of the in-sentence pose according to the utterance speed will be described with reference to FIGS. FIG. 3 is a diagram showing an example of the occurrence frequency for the sentence pause for each utterance speed category (fast, normal, slow). In FIG. 3, when the utterance speed is low, the maximum frequency of occurrence tends to have a longer pause length than other utterance speeds. When the speaking rate is high, the maximum point of occurrence frequency tends to have a shorter pause length than other speaking rates.

  FIG. 4 is a diagram in which the pose length of the in-sentence pose is divided by each utterance speed, and the pose length is shown in units of mora instead of milliseconds. In FIG. 4, even at different utterance speeds, the pose length of the maximum occurrence frequency in the sentence pose is the same in the vicinity of 2 mora. In other words, the pose lengths of the in-sentence poses shown in units of mora are more likely to match regardless of the speaking speed. That is, if the pose length of the in-sentence pose is determined, the speech rate can be calculated from the pose length.

  Next, the tendency of the pause length of the pause between sentences depending on the speech speed will be described with reference to FIG. FIG. 5 is a diagram illustrating an example of an occurrence frequency for a sentence pause for each utterance speed category (fast, normal, or slow). In FIG. 5, when the utterance speed is low, the maximum point of occurrence frequency tends to have a longer pause length than other utterance speeds. When the speech rate is fast, the maximum point of the occurrence frequency distribution tends to have a shorter pause length than other speech rates.

  Here, the difference between each pose length, which is the maximum point with respect to the occurrence frequency for each category of speech speed, is larger in the inter-sentence pose than in the intra-sentence pose (see FIG. 3). For example, comparing the case where the speech rate is fast and the case where the speech rate is slow, the difference between pause lengths, which becomes the maximum point of each occurrence frequency, is greater in the pause between sentences. That is, it can be seen that the pause between sentences shows a tendency of the pause length with respect to the utterance speed more significantly than the pose within the sentence, and is suitable for the purpose of roughly classifying the utterance speed.

  As described above, inter-sentence and intra-sentence poses classified by phrase attributes of speech near the pose have high correlation between the length of each pose and the speech speed, and are useful information for calculating the speech speech speed. is there. Further, since the pause length is the time length of the non-speech interval, it can be accurately detected even for a speech signal in which a signal other than speech is superimposed on speech. Therefore, in the present invention, paying attention to the poses classified by the phrase attributes of the speech near the pose, the speech utterance speed is calculated based on the pose length of the pose.

(First embodiment)
The first embodiment according to the present invention will be described below with reference to FIG. FIG. 6 is a block diagram showing the audio speed conversion device 1 according to the first embodiment of the present invention. In FIG. 6, the voice speed conversion apparatus 1 includes a voice signal storage unit 11, a pause detection unit 12, a statistical data storage unit 13, a speech speed calculation unit 14, a speech speed control unit 15, a speed conversion unit 16, and a speaker 17. Prepare.

  The audio signal storage unit 11 stores an audio signal. Here, the voice signal means a signal including the voice of the speaker who is the target of the speech speed conversion for which the user desires the speech speed conversion. Note that it is only necessary that the voice signal includes at least the voice of the target speaker. That is, the voice signal may be a voice signal composed only of the voice of the target speaker (for example, conversation), or other signals other than the voice may be included in the voice of the target speaker. It may be a superimposed audio signal (for example, an audio signal of a television program, a radio program, a movie recorded on a recording medium, etc.).

  Further, the target speaker is not limited to one speaker, and may be a plurality of speakers. For example, a conversation or the like is composed of a plurality of speaker voices. Then, the voices of the plurality of speakers are all determined as voice sections by the pause detection unit 12 described later.

  Further, the audio signal may be supplied to the audio signal storage unit 11 via a communication medium, for example. Alternatively, an audio signal recorded on a recording medium (for example, an optical disk) may be supplied to the audio signal storage unit 11.

  The pause detection unit 12 determines a speech segment and a non-speech segment from the speech signal stored in the speech signal storage unit 11. Then, the determined non-voice section is detected as a pause. The statistical data storage unit 13 stores statistical data relating to a previously determined pause length. The utterance speed calculation unit 14 calculates the utterance speed based on the pose detected by the pose detection unit 12 and the statistical data stored in the statistical data storage unit 13. The speech rate control unit 15 calculates the compression / expansion rate by a preset control method according to the speech rate calculated by the speech rate calculation unit 14. The speed conversion unit 16 performs speech speed conversion by compressing and expanding the audio signal stored in the audio signal storage unit 11 on the time axis based on the compression / expansion rate calculated by the speech rate control unit 15. Then, the voice signal whose speech speed has been converted is reproduced from the speaker 17. Below, the function of each component in the audio speed conversion apparatus 1 will be described in detail.

  The pause detection unit 12 divides the audio signal stored in the audio signal storage unit 11 into frames (the time length of one frame is a time Tf) set as a detection unit, and an audio section and a non-audio for each frame. Determine the interval. Then, the determined non-voice section is detected as a pause. FIG. 7 is a diagram illustrating an example of a time axis waveform in an audio signal for one frame. In FIG. 7, the vertical axis indicates the level, and the horizontal axis indicates time. In the time axis waveform shown in FIG. 7, the speech sections determined by the pause detection unit 12 are To1 to To6, and the non-speech sections are Tp1 to Tp5.

Here, as a method of determining the speech section and the non-speech section used by the pause detection unit 12, for example, using the Bayes function described in Document 2, the speech section of the target speaker and the speech other than the speech There are known methods for determining other signal sections (Yasuyuki Nakajima, 4 others, "Audio indexing from MPEG encoded data", IEICE Transactions, IEICE, May 2000). D-II Vol.J83-D-II No. 5, p.13
61-1371, hereinafter referred to as Document 2). In the present embodiment, the pause detection unit 12 uses the method of Document 2 above to determine the speech segment of the target speaker from the time axis waveform of the speech signal as a speech segment (for example, speech segments To1 to To6 in FIG. 7). Then, a section of a signal other than the voice is determined as a non-speech section (for example, non-speech sections Tp1 to Tp5 in FIG. 7). Then, the determined non-voice section is detected as a pause.

  In addition, the determination method of an audio | voice area and a non-voice area is not limited to the method of the said literature 2, You may use another method. As an example of another method, signals such as television programs and movies include video signals in addition to audio signals. At this time, using the video signal, video information of a speech organ that moves when the voice of the target speaker is uttered (for example, the lips, jaws, and vocal cords of the target speaker) is recognized. Then, using the recognition result, it is determined whether or not the target speaker is speaking, and if it is speaking, it is determined to be a speech section, and otherwise, it is determined to be a non-speech section, thereby detecting a pause. A method may be used.

  The statistical data storage unit 13 stores pose length statistical data for attributes (hereinafter referred to as classes) of poses obtained in advance. FIG. 8 is a diagram showing an example of statistical data for the pause length of each class. It is assumed that the class is composed of two attributes, an utterance speed attribute and a phrase attribute. Here, the utterance speed attribute indicates an attribute related to the utterance speed classified into fast, normal, and slow. The phrase attribute refers to an attribute relating to a phrase divided between sentences and within a sentence. That is, the class is a combination of two according to the speech speed attribute classification and the phrase attribute classification. Specifically, in FIG. 8, for example, a class is a combination of “in the sentence” that is a category of phrase attributes and “fast” that is a category of speech rate attributes, and a total of 3 (speech rate attributes) * There are 6 classes in total, 2 (word attributes). Then, statistical data of pause length is set for each class. For example, the statistical data is data in which the average value and the standard deviation of the pause length are statistically obtained for each class in advance.

  Note that the statistical data may be prepared based on the speaker's utterance style and characteristics such as whether it is in a reading style or not. Further, statistical data based on TV program genres such as speaker information, sports programs, or dramas may be prepared. At this time, the user may select statistical data according to the content of the audio signal for which the user wants to convert the speech speed, or the optimum statistical data is automatically selected based on EPG information such as TV program broadcast. You may be made to do.

  The speech rate calculation unit 14, for N frames (N is a natural number) before the frame including the current frame detected by the pause detection unit 12, pause length and statistical data storage unit 13 of each pause included in the N frame. Is used to calculate the speech rate of the N frame, and the speech rate of the N frame is set as the speech rate of the current frame. Here, the N frame means the N frames, and a plurality of poses included in the N frames is a pose sequence. For example, when N = 5, a plurality of poses included in 5 frames including the current 1 frame and the 4 frames before the current frame become a pose sequence. Further, for example, when N = 1, a plurality of poses included in one current frame is a pose sequence.

  Also, the time length Tfn of the pause sequence is Tfn = Tf * N because there are N frames of time length Tf. For example, if the time length of the pause sequence is Tfn = 10 seconds and N = 10, the time length of the frame is Tf = 1 second, and the speech rate is calculated from the pause length of each pause included in 10 seconds (Tfn). , The speech rate for the current one second (Tf). In addition, each time quoted above is not a limit value at all. The shorter the time length (Tf) of the frame, the shorter the time interval in which the speech rate is reflected. In addition, as the time length (Tfn) of the pose sequence increases, the number of poses included in the pose sequence increases, and a more accurate speech rate can be calculated. However, if the time length (Tfn) of the pose sequence is too long, Response to changes in the speaking rate of the speaker may be delayed, and the processing burden on the device may increase. Based on the above characteristics, the time length Tfn of the pause sequence and the time length Tf of the frame are appropriately set. FIG. 9 is a block diagram showing the configuration of the speech rate calculation unit 14. The utterance speed calculation unit 14 includes a pose length measurement unit 141, a pose classification unit 142, a pose frequency calculation unit 143, and an utterance speed conversion unit 144.

  In FIG. 9, the pose length measurement unit 141 measures the pose length of each pose with respect to the N frames before the frame including the current frame detected by the pose detection unit 12.

  The pose classification unit 142 classifies the class to which each pose belongs from each pose length constituting the pose string measured by the pose length measurement unit 141. FIG. 10 is a block diagram illustrating a configuration of the pose classification unit 142. The pose classification unit 142 includes a class identification unit 145, a pose string determination unit 146, and a class determination unit 147.

The class identifying unit 145 includes a class (hereinafter referred to as a class to which each pose constituting the pose sequence belongs) from each pose length constituting the pose sequence measured by the pose length measuring unit 141 and the statistical data stored in the statistical data storage unit 13. , Which belongs to the class). Specifically, the degree to which class the pose matches (hereinafter referred to as the matching level L) is calculated for every class for each pose. The matching degree L is calculated by the following formula.
L = 1 / d (1)
d = | x−a | / S (2)
Here, d indicates what is generally called Mahalanobis distance, S indicates the standard deviation of the class, and a indicates the average value of the class. X indicates the pose length of the pose for which the adaptation is calculated. Based on the calculation result of the above equation, the class having the maximum matching level L is identified as the class to which the pose belongs.

  The pose sequence determination unit 146 checks whether the utterance speed attribute classification is the same for all poses constituting the pose sequence among the classes belonging to each pose in the pose sequence identified by the class identification unit 145. . If all the speech rate attribute categories match, the category is determined to be the speech rate category of the pause sequence. On the other hand, if the utterance speed attribute classifications do not partially match, the utterance speed attribute classification of the pose string in which the phrase attribute classification is between sentences in each pose constituting the pose string is the most frequent classification of the pose string. It is determined as the speech rate category. Here, the pause between sentences was used to determine the speech rate classification because the tendency of pause length due to the classification of speech speed attributes is more noticeable in the sentence pause than in the sentence pause. (See FIG. 5).

  The class determination unit 147 re-determines the affiliation class of the pose for the pose whose utterance rate attribute identified by the class identification unit 145 does not match the utterance rate classification determined by the pose sequence determination unit 146. To do. First, the utterance speed attribute classification of the non-matching pose is changed to the utterance speed classification determined by the pose string determination unit 146. Next, the value of the matching level L is compared in the phrase attribute classification (between sentences and within the sentence) that becomes the changed speech speed classification. Then, the classification of the phrase attribute of the class having a large matching level L is determined as the phrase classification of the pose that did not match. As described above, the affiliation class of the non-matching pose is determined. In this way, by using the pose sequence, it is possible to identify poses that are difficult to distinguish only by the length of time, such as whether the pose belongs to the “fast, sentence-to-sentence” or “slow, in-sentence” class. It is possible to accurately determine the belonging class. As described above, the pose classification unit 142 classifies which class each pose belongs to from each pose length constituting the pose string measured by the pose length measurement unit 141.

  The pose frequency calculation unit 143 calculates, for each pose length, the frequency of occurrence of poses whose word / phrase classification is classified as in the sentence among the classes belonging to each pose constituting the pose row classified by the pose classification unit 142. At this time, according to the speech rate attribute determined by the pose sequence determination unit 146 (for example, three categories of fast, normal, and slow), the number of classes and the class width for counting the frequency of occurrence of the pose length of the in-sentence pose are calculated. select. Here, the class number indicates the number by which a predetermined range is divided into sections (classes), and the width of the divided section is referred to as a class width.

  Here, in the present embodiment, the number of classes for counting the frequency of occurrence frequency of the pose length of the in-sentence pose is set as follows. For example, when the speed is fast, the class number is 8 in the time range from 160 milliseconds to less than 220 milliseconds, and when the speed is normal, the class number is 6 in the time range from 230 milliseconds to less than 290 milliseconds, and the case is slow. Is 4 in the time range from 300 milliseconds to less than 360 milliseconds.

  The reason for setting the above time range is that the voice rate of the announcer's voice is fast and around 80 milliseconds per mora. In addition, it is difficult to maintain a natural prosody at a low utterance speed if it exceeds 180 milliseconds per mora. Therefore, in an utterance style in which a sentence is read out independently like an announcer, the utterance speed exists between 80 milliseconds / mora and 180 milliseconds / mora. Here, as shown in FIG. 4, the pose length of the in-sentence pose is statistically about 2 mora. Therefore, the range that can be taken by the pose for conversion to the speech speed using a statistical relational expression in which the pose length of the in-sentence pose is 2 mora is a range from 160 milliseconds to 360 milliseconds. It should be noted that the time range, the number of classes, and the class width for counting the frequency of occurrence of the in-sentence pauses described above are examples and may be set as appropriate.

  The utterance speed conversion unit 144 obtains the pose length that is the maximum frequency among the occurrence frequencies of the in-sentence poses calculated by the pose frequency calculation unit 143, and converts the pose length into the utterance speed. As a method of converting the pose length of the pose in the sentence to the utterance speed, the median value of the class to which the maximum pose length belongs is used to calculate the utterance speed using the above statistical relational expression (length of 2 mora). Convert to. Here, when there are a plurality of classes having the same occurrence frequency, the average value of the pose lengths belonging to the plurality of classes is used for adjacent classes. When a plurality of classes having the same occurrence frequency are not adjacent to each other, the values obtained by adding the occurrence frequencies of each class having the same value and a class adjacent to each class are compared, and the largest class is adopted.

  In the above description, the length of the pose length in the sentence pose is 2 mora as a statistical relational expression, but the present invention is not limited to this. For example, an optimum value may be set for each target speaker. As a result, the speech rate can be converted with higher accuracy. As described above, the speaking rate calculation unit 14 calculates the speaking rate for each pause sequence. Then, the speech rate of the pause sequence is set as the speech rate of the current frame constituting the pause sequence.

  Here, a flow of processing from detection of a pause from an audio signal to calculation of an utterance speed based on the pause length of the pause will be described with reference to FIGS. FIG. 11 is a flowchart showing the flow of processing until the speech rate is calculated based on the pause length. FIG. 12 is a diagram illustrating an example of a calculation result of the matching degree L of each pose constituting the pose row. In FIG. 12, pauses Tp1 to Tp5 correspond to pauses Tp1 to Tp5 in FIG. Further, the calculation result of FIG. 12 is calculated using the statistical data values shown in FIG. 8, and only the current one frame is shown, and the other N frames are omitted. FIG. 13 is a diagram illustrating the results of steps S3 to S6 described later. In order to make the following description more specific, it is assumed that the speech rate of the speaker's voice included in the voice signal stored in the voice signal storage unit 11 is slow, and the “correct answer” column shown in FIG. The affiliation class of each pose based on the utterance speed (slow) is shown.

  First, the pause detection unit 12 (see FIG. 1) determines a voice interval and a non-voice interval for each frame, and detects the non-voice interval as a pause (step S1). Next, the pose length measurement unit 141 measures the pose length of each pose (pose sequence) in the N frames before the frame including the current frame detected by the pose detection unit 12 (step S2), and performs processing. Proceed to the next step.

  Next, the class identifying unit 145 uses the pose lengths measured by the pose length measuring unit 141 and the statistical data stored in the statistical data storage unit 13 using the above formulas (1). ) And equation (2), the matching degree L is calculated for each pose. Then, the class having the maximum adaptability L value in each pose constituting the pose row is identified as the class to which the pose belongs (step S3), and the process proceeds to the next step. Here, in the calculation result of the matching level L shown in FIG. 12, for example, the pose Tp1 becomes the maximum value when the matching level L of the class “sentence between sentences is slow” is “1.2”, and the pose Tp1 belongs to The class is identified as “between sentences, slow”. With the above method, the belonging class is identified for each pose constituting the pose string. The identification results of the pauses Tp1 to Tp5 are the results shown in FIG.

  Next, the pose sequence determination unit 146 determines whether the utterance speed attribute classifications are the same for all poses of the classes belonging to each pose constituting the pose sequence identified by the class identification unit 145 ( Step S4). Then, when the speech rate attribute categories all match, the pause sequence determination unit 146 determines that the category is the speech rate category of the pause sequence, and proceeds to the next step S7. On the other hand, the pose sequence determination unit 146 proceeds to the next step S5 when the utterance speed attribute classifications do not partially match.

  In step S5, the class determination unit 147 has the highest number of utterance speed categories of the pose string among the utterance speed attribute categories of poses in which the phrase attribute classification is between sentences in each pose constituting the pose string. Is determined. Then, the class determination unit 147 performs, for a pose in which the utterance speed attribute classification of the belonging class identified in step S3 and the utterance speed classification determined in step S5 do not match, the belonging class of the pose. Is re-determined (step S6), and the process proceeds to the next step S7. Specifically, the utterance speed attribute classification of the pose that did not match is changed to the utterance speed classification determined in step S5. Next, the matching degree L value is compared in the phrase attribute division (between sentences and within the sentence) that becomes the changed speech speed classification. Then, the classification of the phrase attribute of the class having a large matching level L is determined as the phrase classification of the pose that did not match. As described above, the belonging class of the pose that did not match is determined.

  Here, the flow of steps S3 to S6 will be described in detail with reference to an example of the pose identification result of the pose row where N = 1. In the pose identification result for one frame shown in FIG. 13, the utterance speed attribute classification is “slow” with 4 poses and “fast” pose with one utterance speed attribute classification. They do not match (step S4). Also, in step S5, the phrase speed attribute category of the pause in which the phrase attribute category is “between sentences” is “slow” at three locations and “fast” at one location. Since there are the most “slow” segments, the speech rate category of each pose in the frame is determined to be “slow”.

  Next, since it is determined in step S5 that the speech rate category is “slow”, the pose Tp4 in which the speech rate attribute category is identified as “fast” in step S3 does not match. Therefore, the speech rate attribute category of pause Tp4 is changed to the speech rate category “slow” determined in step S5. Then, in FIG. 12, the value of the matching level L is compared between the phrase attributes “between sentences” and “within sentence” in which the category of the speech rate attribute of pause Tp4 is “slow”. At this time, the value of the adaptation level L of the class “slow, in sentence” is 1.11, and the value of the adaptation level L of the class “slow, between sentences” is 0.41. Accordingly, since the value of the matching level L is larger in the “slow, in sentence” class, the phrase division of the pause Tp4 is determined as “in sentence”. The affiliation class of each pose constituting the determined pose sequence matches the affiliation class (“correct answer” column shown in FIG. 13) based on the actual speech rate of the voice of the speaker. As described above, the affiliation classes of all poses constituting the pose sequence can be matched with the affiliation classes of the respective poses based on the actual speaking rate of the speaker.

  In step S7, the pose frequency calculation unit 143 generates the frequency of occurrence of poses whose word category is determined to be “in-sentence” among the classes belonging to each pose constituting the pose sequence determined in step S4 or step S6. The frequency is calculated for each pause length. Then, the utterance speed conversion unit 144 obtains the pose length that is the maximum frequency among the occurrence frequencies of the in-sentence poses calculated in step S7, and converts it to the utterance speed using a statistical relational expression (step S8). As a result, the utterance speed for each pose string is calculated, and the utterance speed of the pose string is set as the utterance speed of the current frame constituting the pose string. Next, when the process is continued, the process returns to step S1 to continue the process, and when the process is terminated, the process according to the flowchart is terminated (step S9). This is the end of the description of the flow of processing for detecting a pose and calculating the speech rate based on the pose length of the pose.

  Returning to FIG. 6, the speech rate control unit 15 calculates the compression / expansion rate based on a predetermined control method. A control method for reproducing the speech rate of the speaker's voice included in the speech signal at the speech rate desired by the user is set in the speech rate control unit 15.

  As a specific control method, for example, news is uttered at an average speed of 9.5 (mora / second). Therefore, when the calculated speech rate of the pause sequence is higher than 8 (mora / second) so that the speech rate of the current frame is 8 (mora / second), the speech rate of the current frame is 8 ( The compression / expansion rate is controlled so as to extend along the time axis so as to achieve mora / second. If the calculated speech rate of the pose sequence is slower than 8 (mora / second), the compression / expansion rate is compressed on the time axis so that the speech rate of the current frame is 8 (mora / second). There are ways to control.

  Also, there are many topic changes at the beginning of the sentence, and if you miss it, it may be difficult to grasp the contents of the words that follow. Therefore, a control method for controlling the compression / expansion rate may be set in advance so that the phrase after the pause between sentences is slower than the phrase after the other pause.

  The speed conversion unit 16 compresses and decompresses the voice signal stored in the voice signal storage unit 11 on the time axis based on the compression / expansion rate calculated by the speech speed control unit 15, thereby reducing the speed of the speaker's voice. The speech speed is converted to a speed desired by the user from high speed to high speed. Here, the audio signal compression / decompression method includes, for example, the audio speed conversion method disclosed in Japanese Patent No. 3156020, but is not limited to this method, and other methods may be used.

  As described above, the audio speed conversion device described in the first embodiment can detect a pause from an audio signal and accurately calculate the speech rate based on the pause length of the pause. As a result, it is possible to convert the speech speed to the speech speed of the playback voice desired by the user according to the speech speed of the speaker's voice.

  In the above description, the speech rate attribute is classified into three categories “fast, normal, slow”, but may be two categories “fast, slow”. Needless to say, four or more categories such as “Fast, Slightly Fast, Normal, Slightly Slow, Slow” may be used. At this time, each pose constituting the pose sequence is determined as a class belonging to the category of the speech rate attribute, and the speech rate of the speaker's voice can be calculated with a resolution corresponding to the category of the speech rate attribute.

  Note that the audio speed conversion apparatus 1 according to the present embodiment may be realized by causing a general computer system to execute an audio speed conversion program. FIG. 14 is a block diagram illustrating a configuration example in which the audio speed conversion device 1 is realized by the computer system 2. In FIG. 14, the audio signal storage unit 11, the pause detection unit 12, the statistical data storage unit 13, the speech rate calculation unit 14, the speech rate control unit 15, the rate conversion unit 16, and the speaker 17 are each configured in FIG. 1. Since the functions are the same as those of the unit, description thereof is omitted.

  In FIG. 14, the computer system 2 includes a CPU 21, a storage unit 22, and a disk drive device 23. The CPU 21 implements the same functions as the pause detection unit 12, the speech rate calculation unit 14, the speech rate control unit 15, and the speed conversion unit 16 by executing the voice speed conversion program. The storage unit 22 is composed of a recording medium such as a hard disk, and implements the same functions as the audio signal storage unit 11 and the statistical data storage unit 13 described above by executing an audio speed conversion program. The disk drive device 23 reads the audio speed conversion program from the recording medium 24 in which the audio speed conversion program for causing the computer system 2 to function as an audio speed conversion device is stored. When the audio speed conversion program is installed in an arbitrary computer system 2, the computer system 2 can function as the above-described audio speed conversion device. Then, the reproduced sound whose speech speed has been converted by the computer system 2 is reproduced from the speaker 17. The recording medium 24 is a recording medium in a format that can be read by the disk drive device 23 such as a flexible disk or an optical disk. The audio speed conversion program may be installed in the computer system 2 in advance. The speaker 17 may be built in the computer system 2 or may be outside the computer system.

  In the above description, the audio speed conversion program is provided by the recording medium 24, but may be provided by an electric communication line such as the Internet. Further, the processing in the audio speed conversion apparatus may be in a form in which all or a part thereof is processed by hardware such as a speech speed conversion processing device.

(Second Embodiment)
Next, a second embodiment according to the present invention will be described with reference to FIG. FIG. 15 is a block diagram showing an audio speed conversion device 3 according to the second embodiment of the present invention. In FIG. 15, the voice speed conversion device 3 includes a voice signal storage unit 11, a pause detection unit 12, a statistical data storage unit 13, a speech rate calculation unit 14, a speech rate control unit 15, a speed conversion unit 16, a speaker 17, and a speech rate. A storage unit 31, a speed input unit 32, a reproduction time calculation unit 33, and a display unit 34 are provided. The voice signal storage unit 11, the pause detection unit 12, the statistical data storage unit 13, the speech rate calculation unit 14, the speech rate control unit 15, the speed conversion unit 16, and the speaker 17 are the same as those described in the first embodiment. Since it has the same function as the speed converter 1, the same reference symbols are attached and detailed description is omitted.

  First, as described in the first embodiment, the speech rate for each frame is determined by detecting a pause from the speech signal stored in the speech signal storage unit 11 by the pause detection unit 12, and the pause length and statistical data of the pause. Based on the statistical data stored in the storage unit 13, the speech rate calculation unit 14 calculates it.

  The speech rate storage unit 31 stores speech rate data for each frame calculated by the speech rate calculation unit 14 for all data stored in the audio signal storage unit 11. Next, in the speed input unit 32, the utterance speed of the reproduced voice after the speech speed conversion desired by the user is displayed on the screen with an indicator or the like. Then, the user selects or inputs the utterance speed using an input device (not shown).

  The speech rate control unit 15 compresses the compression / expansion rate by a predetermined control method from the speech rate after the speech rate conversion input by the speed input unit 32 and the speech rate of each frame stored in the speech rate storage unit 31. Is calculated. As a predetermined control method, for example, a control method is set such that the utterance speed of each frame approaches the utterance speed input by the speed input unit 32.

  The playback time calculation unit 33 calculates the playback time of the entire speech signal after the speech speed conversion based on the compression / expansion rate for all data of the speech signal calculated by the speech rate control unit 15 and displays it on the display unit 34. Let

  As described above, the audio speed conversion device 3 according to the present embodiment can input the entire audio signal in advance without actually reproducing the audio or image of the audio signal if the speech speed of the playback audio desired by the user is input. You can grasp the playback time.

(Third embodiment)
Next, a second embodiment according to the present invention will be described with reference to FIG. FIG. 15 is a block diagram showing an audio speed conversion device 4 according to the third embodiment of the present invention. In FIG. 15, the voice speed conversion device 4 includes a voice signal storage unit 11, a pause detection unit 12, a statistical data storage unit 13, a speech rate calculation unit 14, a speech rate control unit 15, a speed conversion unit 16, a speaker 17, a speech rate. A storage unit 31, a speed input unit 42, a playback speed calculation unit 43, and a display unit 44 are provided. The voice signal storage unit 11, the pause detection unit 12, the statistical data storage unit 13, the speech rate calculation unit 14, the speech rate control unit 15, the speed conversion unit 16, the speaker 17, and the speech rate storage unit 31 are the first and Since it has the same function as the audio speed conversion apparatuses 1 and 3 described in the second embodiment, the same reference symbols are attached and detailed description is omitted.

  First, as described in the first embodiment, the speech rate for each frame is determined by detecting a pause from the speech signal stored in the speech signal storage unit 11 by the pause detection unit 12, and the pause length and statistical data of the pause. Based on the statistical data stored in the storage unit 13, the speech rate calculation unit 14 calculates it. Next, the speed input unit 32 displays the playback time of the entire speech signal after the speech speed conversion desired by the user on the screen with an indicator or the like. Then, the user selects or inputs the reproduction time of the entire audio signal using the input device.

  The speech rate control unit 15 stores the speech rate storage unit 31 based on the playback time of the entire speech signal after the speech rate conversion input by the time input unit 42 and the playback time of the entire speech signal before the speech rate conversion. Using the utterance speed of each frame, the compression / expansion rate is calculated by a predetermined control method. As a predetermined control method, for example, a control method is set such that the reproduction time of the entire audio signal before the speech speed conversion approaches the reproduction time of the entire audio signal input by the time input unit 42.

  The playback speed calculation unit 43 calculates the speech speed of the playback voice after speech conversion based on the compression / expansion rate for all data of the audio signal calculated by the speech speed control unit 15 and displays the speech speed on the display unit 44.

  As described above, the audio speed conversion device 3 according to the present embodiment converts the speech in advance without actually reproducing the audio or the image of the audio signal if the reproduction time of the entire audio signal desired by the user is input. The utterance speed of later playback voice can be grasped.

  In addition, according to the audio speed conversion device described in the second and third embodiments, the information (the utterance of the reproduced audio) after the speech speed conversion is performed in advance without the user actually reproducing the audio or image of the audio signal. (Speech speed and playback time of the entire audio signal, etc.) can be grasped, so that speech speed conversion more suitable for the user's intention can be provided. Also, by displaying the information after the speech speed conversion, the user can more intuitively perform the speech speed conversion that matches his / her desire.

  An audio speed conversion device, an audio speed conversion method, and an audio speed conversion program according to the present invention can accurately determine an utterance speed of a speaker's voice even if the signal is a signal obtained by superimposing a signal other than the voice on the speaker's voice. It can be calculated and applied to uses such as a speech reproduction device, speech recognition device, speech summarization device and the like that perform speech speed conversion.

The figure which shows an example of the occurrence frequency of the pause according to pause length according to the classification of speech speed (fast, normal, slow) Schematic diagram showing an example of a pause in the time axis waveform of speech The figure which shows an example of the occurrence frequency of the pause in a sentence by the classification of utterance speed (fast, normal, and slow) The figure which showed the pose length of the sentence pose in mora The figure which shows an example of the occurrence frequency of the pause between sentences by the classification of speech speed (fast, normal, slow) The block diagram which shows the audio | voice speed converter 1 which concerns on the 1st Embodiment of this invention. The figure which shows an example of the time-axis waveform of the audio | voice signal for 1 frame The figure which shows an example of the statistical data of the pose length of each class The block diagram which shows the structure of the speech rate calculation part 14 Block diagram showing the configuration of the pose classification unit 142 The figure which shows the flowchart showing the flow of a process until utterance speed is calculated based on pause length. The figure which shows an example of the calculation result of the conformity degree L of each pose which comprises a pose row | line | column. The figure which shows each result of step S3-S6 The block diagram which shows the structure by which the audio | voice speed converter 1 is implement | achieved by the computer system 2 The block diagram which shows the audio | voice speed converter 3 which concerns on the 2nd Embodiment of this invention. The block diagram which shows the audio | voice speed converter 4 which concerns on the 3rd Embodiment of this invention. A block diagram showing a configuration in which a conventional audio compression / decompression apparatus is applied to an IC recorder

Explanation of symbols

1, 3, 4 Audio speed conversion device 11 Audio signal storage unit 12 Pause detection unit 13 Statistical data storage unit 14 Speech rate calculation unit 15 Speech rate control unit 16 Speed conversion unit 17 Speaker 141 Pause length measurement unit 142 Pause classification unit 143 Pause Frequency calculation unit 144 Speech rate conversion unit 145 Class identification unit 146 Pause sequence determination unit 147 Class determination unit 2 Computer system 21 CPU
22 storage unit 23 disk drive device 24 recording medium 31 utterance speed storage unit 32 speed input unit 33 reproduction time calculation unit 34 display unit 42 speed input unit 43 reproduction speed calculation unit 44 display unit

Claims (12)

An audio speed conversion device that converts an audio signal including a voice of a speaker to be converted into an audio speed and reproduces the audio signal,
A non-speech section detecting unit that distinguishes a speech section in which the voice of the speaker is included from the speech signal and a non-speech section in which the speech of the speaker is not included, and detects the non-speech section;
A non-speech section length measuring unit that measures a time length for each non-speech section detected from the speech signal by the non-speech section detector;
Based on the time length of each non-speech segment measured by the non-speech segment length measurement unit, an utterance rate calculation unit that calculates the speech rate of the speaker in the speech signal;
A speech speed conversion device comprising: a speech speed conversion / playback section that converts the speech speed according to the speech speed calculated by the speech speed calculation section and plays back the speech signal. The speech rate calculation unit includes an occurrence frequency calculation unit that calculates an occurrence frequency of a time length for a non-speech interval measured by the non-speech interval length measurement unit,
The utterance speed calculation unit is configured to determine the utterance of the speaker according to the time length that is the maximum frequency in the occurrence frequency, based on a statistical relational expression between the utterance speed obtained in advance and the time length of the non-speech interval. The speed conversion apparatus according to claim 1, wherein the speed is calculated. The voice speed conversion device further includes a statistical data storage unit that stores statistical data of a time length of the non-voice section statistically obtained for each of a plurality of preset segments,
The speech rate calculation unit is configured to classify the non-speech section according to the time length of the non-speech section measured by the non-speech section length measurement unit based on the statistical data stored in the statistical data storage unit. A non-speech segment classification unit that classifies each segment and selects one segment from the plurality of segments based on a predetermined condition;
The speech velocity conversion apparatus according to claim 2, wherein the occurrence frequency calculation unit calculates the occurrence frequency using a time length of a non-speech section belonging to the category selected by the non-speech segment classification unit. . The statistical data storage unit sets the plurality of categories according to speech rates classified into a plurality of categories, and stores the statistical data for each of the categories,
The non-speech segment classification unit is configured to classify the non-speech segment according to the time length of the non-speech segment measured by the non-speech segment length measurement unit based on the statistical data for each segment into which the speech speed is classified. 4. The voice speed conversion device according to claim 3, wherein the voice speed conversion device is classified for each of the categories, and a category in which the non-voice segment is classified most frequently among the categories is selected. The statistical data storage unit statistically obtained the statistical data for the first section obtained statistically the time length of the non-speech interval occurring immediately after the reading and the non-speech interval time length produced immediately after the punctuation And statistical data for the second category,
The non-speech segment classification unit is configured to perform the non-speech segment according to the time length of the non-speech segment measured by the non-speech segment length measurement unit based on statistical data for each of the first segment and the second segment. And classifying each of the categories and selecting the first category,
The speech velocity conversion device according to claim 3, wherein the occurrence frequency calculation unit calculates the occurrence frequency using a time length of a non-speech interval belonging to the first category. The statistical data storage unit sets the plurality of categories by a combination of a plurality of classifications of speech speeds, a first category that occurs immediately after reading, and a second category that occurs immediately after a phrase. Is stored statistical data of the length of the non-speech interval obtained statistically,
The non-speech segment classification unit classifies the non-speech segment for each segment according to the time length of the non-speech segment measured by the non-speech segment length measurement unit based on the statistical data for each of the plurality of segments. , By extracting the most classified classification of the non-speech segment from the combination of the plurality of classified classifications according to the speaking speed and the second classification, and determining the classification for the speaking speed, 4. The voice speed conversion apparatus according to claim 3, wherein a classification that is a combination of the classification for the uttered speech speed and the first classification is selected.   The statistical data storage unit sets the plurality of sections in advance according to speaker characteristics, and stores statistical data of the time length of the non-speech section statistically obtained for each of the plurality of sections. The voice speed conversion device according to claim 3, wherein: The speech rate calculation unit
A non-speech section that classifies the non-speech sections detected by the non-speech section detection unit into a plurality of groups using the time length measured by the non-speech section length measurement unit, and selects one from the plurality of groups A classification section;
An occurrence frequency calculating unit that calculates an occurrence frequency using a time length of a non-speech segment belonging to the group selected by the non-speech segment classification unit;
Based on a statistical relational expression between an utterance speed obtained in advance and a time length of the non-speech interval, an utterance speed conversion that calculates the utterance speed of the speaker according to the time length that is the maximum frequency in the occurrence frequency The audio speed conversion device according to claim 1, further comprising: A display unit;
The speech speed conversion / playback unit further includes a playback time calculation unit that calculates a playback time for playback by converting the voice speed of the audio signal and displays information indicating the playback time on the display unit. The audio speed conversion device described in 1. A display unit;
The speech speed conversion / playback unit further includes a playback speed calculation unit that calculates a playback speed at which the audio signal is played back by converting the speech speed and displays information indicating the playback speed on the display unit. The audio speed conversion device described in 1. A voice speed conversion method for converting a voice signal including a voice of a speaker to be converted into a voice speed and reproducing the voice signal,
A non-speech section detecting step for distinguishing between a speech section in which the speech of the speaker is included from the speech signal and a non-speech section in which the speech of the speaker is not included, and detecting the non-speech section;
A non-speech interval length measuring step for measuring a time length for each non-speech interval detected from the speech signal for a predetermined time by the non-speech interval detection step;
Based on the time length of each non-speech segment measured in the non-speech segment length measurement step, an utterance rate calculation step of calculating the utterance rate of the speaker in the speech signal;
A speech speed conversion method comprising: a speech speed conversion reproduction step of reproducing the speech signal by converting the speech speed according to the speech speed calculated in the speech speed calculation step. An audio speed conversion program executed by a computer of an audio speed conversion apparatus that converts an audio signal including a voice of a speaker to be converted into an audio speed and reproduces the audio signal,
In the computer,
A non-speech section detecting step for distinguishing between a speech section in which the speech of the speaker is included from the speech signal and a non-speech section in which the speech of the speaker is not included, and detecting the non-speech section;
A non-speech interval length measuring step for measuring a time length for each non-speech interval detected from the speech signal for a predetermined time by the non-speech interval detection step;
Based on the time length of each non-speech segment measured in the non-speech segment length measurement step, an utterance rate calculation step of calculating the utterance rate of the speaker in the speech signal;
A speech speed conversion program for executing a speech speed conversion / reproduction step of performing speech speed conversion on the speech signal according to the speech speed calculated in the speech speed calculation step.

Images