JP5874341B2 - Audio signal processing apparatus and program - Google Patents

Description

  The present invention relates to an audio signal processing device and a program, and can be applied to, for example, a speech speed conversion device that changes the speaking speed while maintaining the voice quality of a speaker.

  Conventionally, in speech signal processing (speech speed conversion processing) that changes the speaking speed while maintaining the voice quality of the speaker, conventionally, PICOLA (Pointer Interval Control) that expands and compresses on the time axis using the periodicity included in the speech. There is a technique called “Overlap and Add” (see Non-Patent Document 1). PICOLA generates a speech waveform that can smoothly connect the pitch and the subsequent waveform from the pitch (waveform of the speech signal of the basic period) extracted from the speech data, and further inserts the generated speech waveform This is a technique for compressing or expanding the original sound by outputting a predetermined data length. In the following, the conventional PICOLA algorithm will be described in the case where the speech speed is slowed down (the voice is expanded).

  In PICOLA, in order to expand the voice to an arbitrary length, a pitch that is a periodic waveform that repeatedly appears in the voice is used. The pitch is extracted from speech data (hereinafter referred to as “pitch search range TP”) during a period for searching for a predetermined pitch of the speech data using an autocorrelation function or the like. Using the extracted pitch, as shown in FIG. 12, first, waveforms A ′ and B ′ are generated by fading in the waveform A having the pitch W and fading out the waveform B having the same length as the subsequent waveform A. By adding (synthesizing) the waveforms A 'and B', an expansion waveform C that is smoothly connected to the waveforms A and B is generated. By inserting this waveform C between waveform A and waveform B, the speech waveform can be expanded. Since the length of the inserted waveform C is the same as the length of the pitch W (hereinafter also referred to as “Tw”), if the length of the original audio signal is L, the length after expansion is L + Tw. . When the desired elongation rate is Rs, Rs = (L + Tw) / L because the length after the elongation should be R times the original length. Accordingly, the length of the original signal necessary to satisfy the expansion rate Rs is L = Tw / (Rs−1). For example, if the elongation rate Rs is 1.25 times and the pitch length W is 100 samples, L = 400 samples. If the generated 100 samples are added to the original 400 samples and output, the expansion ratio of 1.25 times can be realized by (output sample number 500) / (original signal sample number 400).

FIG. 13 shows a cycle of extension processing by PICOLA. Extracting pitch W ₁ after obtaining the pitch search range Tp of data first generates a speech waveform from the pitch, outputs a frequency of the audio data required for a desired elongation. Last point P ₁ that has output the audio data becomes the start point of the next process. Again extracted pitch Once you have more audio data pitch search range Tp counted from P _1, executes the decompression process. By repeating the above processing, the speech speed can be converted to an arbitrary speed.

  Normally, when processing and output are performed in units of frames with a preset number of samples n as one frame, in order to execute speech speed conversion in real time, data for one frame is obtained every time one frame has elapsed. Processing must be complete. In the following, it is assumed that the talk side conversion device converts voice data in real time and outputs the frame as an “output frame”. Then, in the talk-side conversion apparatus that performs processing in real time in units of frames, the processed audio data for one output frame is held in a buffer (hereinafter referred to as “output buffer”) every time an output frame has elapsed. Must be. However, the speech rate conversion device can output an output frame if the output buffer does not hold processed audio data for one output frame even when the output frame is to be output. In other words, the audio signal is interrupted on the supply side of the output frame and causes an abnormality in the subsequent processing.

  In a speech side conversion device that uses PICOLA as an algorithm for speech speed conversion (when the speech speed is converted slowly), even if it is time to output an output frame, processed speech data for one output frame is stored in the output buffer. Situations that cannot be maintained include the following two types of conditions.

  As a first condition, there is a case where a conventional speech speed conversion apparatus performs an expansion process after pitch extraction from unprocessed audio data. Specifically, in the conventional speech speed conversion apparatus, the audio data before processing is not buffered for the pitch search range Tp, and further, the audio data in the output buffer (audio data after decompression processing) is output by one output. This is a case where the timing for outputting the next output frame has arrived with only the remaining frames or less remaining.

  As a second condition, in a conventional speech speed conversion device, pitch candidates are extracted from unprocessed audio data, but there is a case where the expansion processing is not performed based on the pitch determination criterion. The input voice signal does not always include the pitch of the voice, and in order to convert the speech speed naturally, the vowels and voiced consonants must be expanded to avoid unvoiced and non-voiced sections. . Therefore, in the conventional speech speed conversion device, it is necessary to determine whether or not the pitch candidates obtained as a result of the pitch search are appropriate as the pitch used for the decompression process. In a conventional speech speed conversion device, for example, when a predetermined range of voice data from the beginning is extracted as a pitch candidate, and it is further determined that the pitch candidate is not suitable as a pitch to be used for the expansion process, the voice of the pitch candidate The data is output as it is without being decompressed (held in the output buffer). In a speech speed converting apparatus that outputs voice data in units of frames, the data length of voice data that is a pitch candidate may be shorter than the data length of an output frame. In that case, in the conventional speech speed conversion device, when only the voice data of the pitch candidate that was not used for the decompression process is held in the output buffer, when the timing to output the next frame comes, the output frame is normal Can not be output to.

  Patent Document 1 describes a solution to the problem of the conventional speech speed conversion device as described above. In the device described in Patent Document 1, after the expansion processing by PICOLA is completed, even if there is no data for the pitch search range Tp in the buffer, the pitch information of the previous processing is reused and the expansion processing is performed once again in the same frame. Run. Japanese Patent Application Laid-Open No. H10-228561 describes that this can avoid the generation of a frame without output data.

JP 2006-3517 A

Naotaka Morita and Fumada Itakura, “Expansion and compression of speech over time using pointer movement control (PICOLA) and its evaluation”, Proceedings of the Acoustical Society of Japan, p. 149-150, October 1986

  However, the technique described in Patent Document 1 has a problem that the speech speed conversion cannot be stably operated under any environment.

  Specifically, the drawbacks of the technology described in Patent Document 1 are roughly divided into the following three.

  First, since the technique described in Patent Document 1 requires at least one extension process, there is a problem that it cannot be handled at all when the extension process is not performed after the pitch candidate extraction.

  Secondly, the technique described in Patent Document 1 has a problem that when the decompression process is performed again in the same frame, there is not always enough data remaining in the output buffer for the decompression process. In other words, in the technique described in Patent Document 1, the technique described in Patent Document 1 cannot be applied unless data equal to or larger than the data output in the first decompression process remains in the output buffer.

  Thirdly, in the technique described in Patent Document 1, since the pitch information of the portion not directly related to the data is used at the second expansion within the same frame, the portion that is not directly related to the data is expanded, and the sound quality deteriorates. There is a problem that it ends up.

  In view of the above-described problems, an audio signal processing apparatus and program capable of stably outputting audio data after speech speed conversion processing when performing audio speed conversion processing on audio data of an audio signal in real time Is desired.

The audio signal processing apparatus of the first phone invention includes (1) an input buffer means for storing audio data of an input audio signal, and (2) an audio signal waveform based on the audio data stored in the input buffer means. From the speech signal waveform for the search cycle, extract the fundamental cycle, and using the extracted speech signal waveform of the fundamental cycle, the speech rate conversion means for performing speech rate conversion processing on the speech data stored in the input buffer means; (3) output buffer means for storing voice data after the speech speed conversion means performs the speech speed conversion process; and (4) for the output interval of the voice data stored in the output buffer means for each output interval. and the audio data outputting means for outputting an output audio data frames containing audio data, (5) the input buffer means, by adding the above-mentioned search cycle and the output interval, further, the output From the lowest accumulation period Minimum accumulation period or more audio data obtained by subtracting one sampling period of the audio data is accumulated in Ffa it means, that it has a conversion processing control means for starting the speech speed conversion processing by the speech speed converting means Features.

The audio signal processing program according to the second aspect of the present invention comprises: (1) input buffer means for storing audio data of an input audio signal; and (2) an audio signal based on the audio data stored in the input buffer means. For the waveform, extract the fundamental period from the speech signal waveform for the search period, and use the extracted speech signal waveform of the fundamental period to perform speech speed conversion processing on the speech data stored in the input buffer means Means, (3) output buffer means for storing the speech data after the speech speed conversion means has been subjected to speech speed conversion processing, and (4) of the speech data stored in the output buffer means for each output interval and the audio data outputting means for outputting an output audio data frames containing audio data output interval fraction, and (5) to the input buffer means, the search cycle and the output interval Calculated and, further, the output from the minimum accumulation period Minimum accumulation period or more audio data obtained by subtracting one sampling period of audio data in the buffer means is accumulated, the conversion processing control for starting the speech speed conversion processing by the speech speed converting means It is made to function as a means.

  According to the present invention, it is possible to provide an audio signal processing device capable of stably outputting audio data after speech speed conversion processing when performing speech speed conversion processing on audio data of an audio signal in real time. .

It is a block diagram which shows the functional structure of the speech-speed converter which concerns on 1st Embodiment. It is explanatory drawing explaining the process example of the pitch extraction process with the speech-speed converter which concerns on 1st Embodiment. It is the flowchart shown about the process which outputs an output frame with the speech-speed converter which concerns on 1st Embodiment. 4 is a flowchart (part 1) illustrating an operation from acquisition of an input frame to supply of audio data to an output buffer in the speech speed conversion apparatus according to the first embodiment. 4 is a flowchart (part 2) illustrating an operation from acquisition of an input frame to supply of audio data to an output buffer in the speech speed conversion apparatus according to the first embodiment. 5 is a timing chart showing a first example of an operation when the minimum delay amount is made equal to the pitch search range in the speech speed converting apparatus according to the first embodiment. 6 is a timing chart showing a second example of the operation when the minimum delay amount is made equal to the pitch search range in the speech speed conversion device according to the first embodiment. 6 is a timing chart showing a first example of an operation when the minimum delay amount is set longer than the pitch search range in the speech speed converting apparatus according to the first embodiment. 6 is a timing chart showing a second example of the operation when the minimum delay amount is set longer than the pitch search range in the speech speed converting apparatus according to the first embodiment. It is a block diagram which shows the functional structure of the speech-speed converter which concerns on 2nd Embodiment. 6 is a flowchart illustrating an operation from acquisition of an input frame to supply of audio data to an output buffer in the speech speed conversion apparatus according to the second embodiment. It is explanatory drawing shown about the waveform expansion | extension by PICOLA performed with the conventional speech speed converter. It is a figure for demonstrating the cycle of the expansion | extension process by PICOLA performed with the conventional speech speed converter.

(A) First Embodiment Hereinafter, a first embodiment of an audio signal processing device and a program according to the present invention will be described in detail with reference to the drawings. Note that the audio signal processing device according to the first embodiment is a speech rate conversion device.

(A-1) Configuration of the First Embodiment FIG. 1 is a block diagram showing a functional configuration of the speech rate conversion apparatus 100 of this embodiment.

  The speech speed conversion apparatus 100 includes a data input unit 101, a speech speed conversion processing buffer 102, a delay control unit 103, a speech speed control unit 104, a pitch extraction unit 105, a PICOLA processing unit 106, an output buffer 107, and a data output unit 108. Have.

The speech speed conversion apparatus 100 is an information processing apparatus having a program execution configuration such as a CPU, ROM, RAM, EEPROM, hard disk and the like (not limited to one, but capable of distributed processing of a plurality of units). May be constructed by installing the audio signal processing program or the like of the embodiment, and even in that case, it can be functionally shown as in FIG.
The data input unit 101 generates voice data (for example, voice data generated in a PCM (Pulse Code Modulation) format) sampled at a predetermined sample interval Ts from an input voice signal (acoustic signal) that has been input. Then, the sample number n is divided into frame units, and the divided number is supplied to the speech speed conversion processing buffer 102. Hereinafter, a frame generated by the data input unit 101 is referred to as an “input frame”.

  In this embodiment, as described above, the data input unit 101 is described as encoding the input voice signal to generate an input frame and supplying the input frame to the speech speed conversion processing buffer 102. There is no limitation on the method and method for holding the audio data supplied to the speed conversion processing buffer 102. For example, the data input unit 101 may supply audio data to the speech rate conversion processing buffer 102 not in units of input frames but in units of samples. The data input unit 101 may hold, for example, already encoded audio data from an external device (the amount of data held at one time is not limited).

  The speech speed conversion processing buffer 102 holds the input frame supplied from the data input unit 101, and converts the audio data of the held input frame according to the control of the delay control unit 103, the speech speed control unit 104, the pitch extraction unit 105, and the like. The processing such as supplying to the subsequent processing configuration (PICOLA processing unit 106) is performed.

  The delay control unit 103 controls the operation of other processing configurations (speech speed conversion processing buffer 102, pitch extraction unit 105, etc.) according to the amount of audio data held in the speech speed conversion processing buffer 102. It is. The delay control unit 103 checks whether or not the speech speed conversion processing buffer 102 holds audio data equal to or more than the minimum delay amount Td (minimum accumulation period) immediately after the processing of the speech speed conversion device 100 is started. The delay control unit 103 starts the expansion processing by the PICOLA processing unit 106 or the like when it is confirmed in the speech speed conversion processing buffer 102 that audio data for a period equal to or longer than the minimum delay amount Td is held. In addition, other processing configurations are controlled.

  The minimum delay amount Td needs to be at least a period equal to or greater than the pitch search range Tp. In this embodiment, it is assumed that the minimum delay amount Td is calculated as in the following equation (1). On the other hand, although the details are dictated, the speech speed control unit 104 calculates the data length (period corresponding to the voice data) necessary to achieve the expansion rate from the expansion rate Rs set at the start.

Minimum delay amount Td = pitch search range Tp
+1 output frame period-1 sample period (1)
The pitch extraction unit 105 searches for the pitch of the audio data held in the speech speed conversion processing buffer 102. The pitch extraction unit 105 searches (extracts) pitch candidates for the speech data held in the speech speed conversion processing buffer 102 when speech data for the pitch search range Tp is in the speech speed conversion processing buffer 102. I do.

  For example, the pitch extraction unit 105 selects a pitch candidate having the strongest periodicity from the pitch search range Tp of the speech data held in the speech speed conversion processing buffer 102, and the pitch candidate is the PICOLA processing unit 106. It is also determined whether or not it is suitable for use in speech speed conversion processing. Here, for simplification of description, the pitch candidates extracted by the pitch extraction unit 105 are audio data starting from the beginning (the top in time series) of the audio data held in the speech speed conversion processing buffer 102. And

  Then, the pitch extraction unit 105 determines whether the extracted pitch candidate speech data (speech speed conversion processing buffer 102 speech data within a predetermined range from the top in the time series) is appropriate for use in the speech speed conversion processing. If the speech rate conversion processing buffer 102 is determined to be appropriate, the voice data of the extracted pitch and the length necessary for the speech side conversion processing (decompression processing) following the voice data of the pitch are controlled. The audio data (hereinafter referred to as “decompression processing data period Tg”) is supplied to the PICOLA processing unit 106.

  The length of the expansion processing data period Tg is determined by the speech speed control unit 104 according to the expansion rate Rs.

  On the other hand, when it is determined that the extracted speech data of the pitch candidate is not suitable for use in the speech speed conversion process, the pitch extraction unit 105 controls the speech speed conversion processing buffer 102 to determine the pitch candidate. The audio data is supplied to the output buffer 107.

  The pitch candidate extraction process by the pitch extraction unit 105 may use, for example, a degree of difference or an autocorrelation coefficient. Hereinafter, an example of processing in which the pitch extracting unit 105 extracts pitch candidates using the above-described difference will be described.

For example, as shown in FIG. 2, the pitch extraction unit 105 performs the adjacent data intervals while shifting one sample at a time for the top voice data (“pitch candidate interval” in FIG. 2) of the speech speed conversion processing buffer 102. The degree of difference from the “comparison target section” in FIG. 2 is calculated, and the one with the smallest degree of difference is set as a pitch candidate. For example, the pitch extraction unit 105 may calculate the degree of difference between the temporary pitch candidate section and the comparison target section using the following equation (2). In the following equation (2), f represents the temporary pitch candidate section shown in FIG. 2, and f _i represents the i-th sample value from the beginning of f (pitch candidate section). In the following equation (2), g represents the comparison target section shown in FIG. 2, and g _i represents the i-th sample value from the beginning of g (comparison target section).

  The number of times the pitch extraction unit 105 performs a pitch search (the number of times of calculating the dissimilarity) is not limited. For example, the dissimilarity is calculated up to a predetermined maximum number of times, and the pitch having the smallest dissimilarity is pitched. You may make it extract as a candidate area. Then, when the degree of difference is less than a predetermined threshold, the pitch extraction unit 105 may determine that the pitch candidate is appropriate for use in the expansion process.

  As the pitch search range Tp, for example, it is desirable to set a range sufficient for searching for a human pitch. For example, when the sampling frequency is 8 kHz, the range may be 20 to 120 samples.

  The PICOLA processing unit 106 generates audio data used to expand the audio data held in the speech speed conversion processing buffer 102 based on the audio data (speech waveform) of the pitch W extracted by the pitch extraction unit 105. For example, the PICOLA processing unit 106 may generate audio data to be used for the decompression process using a cross fade, as in the conventional technique.

  Then, the PICOLA processing unit 106 transmits the expansion processing data period Tg according to the length of the waveform of the pitch W (the length of the waveform on the time axis, which is the same as the length of the waveform of the generated audio data). The speed control unit 104 is inquired and acquired.

  Then, when the PICOLA processing unit 106 acquires the expansion processing data period Tg, the PICOLA processing unit 106 acquires audio data for the expansion processing data period Tg following the pitch W among the audio data held in the speech speed conversion processing buffer 102. . Then, the PICOLA processing unit 106 supplies the output buffer 107 with audio data in which the generated audio data is inserted between the acquired audio data of the pitch W and the audio data of the decompression processing data period Tg.

  The PICOLA processing unit 106, when there is no audio data for the decompression processing data period Tg remaining in the speech rate conversion processing buffer 102, is the length of the shortage (hereinafter referred to as “unprocessed data period Tu”). To the speech speed control unit 104.

  The speech speed control unit 104 holds the parameter of the reference expansion rate Rs, and the expansion processing data for satisfying the expansion rate Rs according to the inquiry from the PICOLA processing unit 106 based on the stored expansion rate Rs. The length of the period Tg is calculated and returned. The method by which the speech speed control unit 104 holds the expansion rate Rs is not limited. However, for example, it may be set in advance or may be configured to change according to the user's operation.

  When the speech speed control unit 104 is notified of the length Tw of the pitch W from the PICOLA processing unit 106, the speech speed control unit 104 sets an expansion processing data period Tg corresponding to the length Tw of the pitch W to satisfy the expansion rate Rs. calculate. Then, the speech speed control unit 104 returns the calculated decompression processing data period Tg to the PICOLA processing unit 106.

  The speech speed control unit 104 may calculate the decompression processing data period Tg using, for example, the following equation (3).

Tg = (Tw / (Rs-1))-Tw (3)
When the unprocessed data period Tu is reported from the PICOLA processing unit 106, the speech speed control unit 104 controls the speech speed conversion processing buffer 102 and outputs data for the unprocessed data period Tu audio data. The processing to be supplied to the unit 108 is prioritized. That is, when the unprocessed data period Tu occurs, the speech speed control unit 104 controls the speech speed conversion processing buffer 102 when the speech data for the unprocessed data period Tu is accumulated in the speech speed conversion processing buffer 102 next time. Thus, the audio data is supplied to the output buffer 107 in preference to the next decompression process.

  The output buffer 107 supplies audio data for one output frame to the data output unit 108 every time when an output frame should be transmitted. When the output frame is composed of audio data for n samples, the timing at which the output frame should be transmitted comes every sample interval Ts × n.

  When the voice data held in the output buffer 107 is less than one output frame, such as immediately after the operation of the speech speed converting apparatus 100 is started, the output buffer 107 is prepared with, for example, silence data or previously prepared. An output frame including dummy audio data such as noise may be supplied to the data output unit 108.

  The data output unit 108 outputs the audio data of the output frame supplied from the output buffer 107 by a predetermined method. The method for outputting the audio data by the data output unit 108 is not limited. For example, the audio output device 108 is provided with an audio output device such as a speaker, or is output to a predetermined data storage medium (for example, a hard disk drive). The audio data of the output frame may be stored as it is or may be encoded and inserted into a packet of a predetermined format, and sent to the destination communication device.

(A-2) Operation of the First Embodiment Next, the operation of the speech rate conversion apparatus 100 of the first embodiment having the above configuration will be described.

  First, an output data frame output process by the data output unit 108 in the speech speed conversion apparatus 100 will be described with reference to the flowchart of FIG.

  When speech speed conversion processing (processing for slowing down the speech speed) is started in the speech speed conversion apparatus 100, the output buffer 107 first waits until the next output frame transmission timing arrives (S101), and outputs the output frame. When the timing arrives, audio data for one output frame is supplied to the data output unit 108 (S102). As described above, the timing at which the output buffer 107 should output the output frame arrives every sample interval Ts × n.

  In the output buffer 107, immediately after the speech speed conversion processing is started in the speech speed converting apparatus 100, the audio data used for the output frame is not held, but the audio data is accumulated more than a predetermined amount (for example, one output frame or more). For example, the output frame of the output frame is suspended, or the output frame including dummy audio data (for example, silence data or audio data such as noise prepared in advance) is sent to the data output unit 108. You may make it supply. The output buffer 107 also outputs an output frame including dummy audio data (to the output buffer 107) even when the supply of audio data is started and the audio data to be output is less than one output frame. An output frame in which the remaining audio data is stored as dummy audio data may be supplied to the data output unit 108.

  When the speech speed conversion process is continued in the speech speed conversion apparatus 100, the output buffer 107 returns to the above-described step S101 and waits until the next output frame transmission timing. On the other hand, when the speech speed conversion process is continued in the speech speed conversion apparatus 100, the output buffer 107 ends the process.

  Next, in the speech speed conversion apparatus 100, the process from decompressing the input speech signal (speech data) (processing to slow down the speech speed) and supplying it to the output buffer 107 will be described with reference to FIGS. I will explain.

  When the speech speed conversion processing (processing for slowing down the speech speed) is started in the speech speed conversion apparatus 100, the data input unit 101 generates speech data (for example, PCM format data) from the input speech signal. The generated audio data is divided and acquired for each input frame (S201).

  In the data input unit 101, when an input frame is acquired, the input frame is supplied to the speech speed conversion processing buffer 102 (S202).

  In this embodiment, in order to simplify the description, it is assumed that the sampling interval and the number of samples of the input frame and the output frame are the same. In the following description, it is assumed that the sampling interval between the input frame and the output frame is Ts, and the number of samples is n.

  Then, the delay control unit 103 checks whether or not data having a minimum delay amount Td or more has been held in the speech speed conversion processing buffer 102 even once in the past since the start of the current speech speed conversion process. (S203). When it is confirmed by the delay control unit 103 that no data of the minimum delay amount Td or more has been held in the speech speed conversion processing buffer 102, the speech speed conversion device 100 returns to the above-described step S201 to operate. . On the other hand, if it is confirmed by the delay control unit 103 that data of the minimum delay amount Td or more has been held in the speech speed conversion processing buffer 102, the process proceeds to step S204 described later.

  Then, the speech speed control unit 104 confirms whether or not a setting change related to the speech speed currently applied to the speech speed conversion process has been performed by a user operation or the like, and when it is confirmed that the setting change has been performed. In step S204 and S205, the expansion rate Rs applied to the speech speed conversion process is changed to a corresponding value after the change.

  The speech speed control unit 104 and the speech speed control unit 104 have previously reported the unprocessed data period Tu from the PICOLA processing unit 106, and the unprocessed data period Tu for satisfying the expansion rate Rs. A check is made as to whether or not the supply of the audio data to the output buffer 107 has been completed (that is, whether or not the expansion rate Rs is satisfied) (S206).

  When the speech speed control unit 104 confirms that the expansion rate Rs is not satisfied, the speech speed control unit 104 controls the speech speed conversion processing buffer 102 to perform an unprocessed data period Tu. Audio data is supplied to the output buffer 107 (if the unprocessed data period Tu is not reached, audio data that can be supplied at this time is supplied to the output buffer 107) (S207). Then, the speech speed conversion apparatus 100 operates from step S214 described later.

  On the other hand, when the speech speed control unit 104 confirms that the expansion rate Rs is satisfied, the pitch extraction unit 105 performs speech data corresponding to the pitch search range Tp (the amount required for pitch search). It is confirmed whether or not (voice data) is held in the speech speed conversion processing buffer 102 (S208).

  When the pitch extraction unit 105 determines that the speech data for the pitch search range Tp is not held in the speech speed conversion processing buffer 102, the speech speed conversion device 100 performs the process in step S201 described above. Returning to FIG. 3, the process starts from the process of acquiring the next input frame.

  On the other hand, if the pitch extraction unit 105 determines that the speech data for the pitch search range Tp is held in the speech speed conversion processing buffer 102, the pitch extraction unit 105 extracts pitch candidates (S209). Then, it is determined whether or not the pitch candidate is suitable for use in the extension process (S210).

  If it is determined in step S210 that the pitch candidate is not suitable for use in the decompression process, the pitch extraction unit 105 controls the speech speed conversion processing buffer 102 to obtain the speech data of the pitch candidate. The data is directly supplied to the output buffer 107 (S211). Then, the speech speed conversion apparatus 100 operates from the process of step S214 described later.

  On the other hand, if it is determined in step S210 that the pitch candidate is suitable for use in the decompression process, the pitch extraction unit 105 controls the speech speed conversion processing buffer 102 to output the voice data of the pitch candidate. Are supplied to the PICOLA processing unit 106 as audio data of pitch W. Then, the speech speed conversion apparatus 100 operates from the process of step S212 described later.

  When the voice data having the pitch W is supplied, the PICOLA processing unit 106 uses the voice data (pitch) stored in the speech speed conversion processing buffer 102 based on the voice data (voice waveform) having the pitch W. Audio data having the same length as W) is generated (S212).

  Then, the PICOLA processing unit 106 inquires the speech speed control unit 104 to acquire the expansion processing data period Tg corresponding to the waveform length Tw of the pitch W. Then, when the PICOLA processing unit 106 acquires the expansion processing data period Tg, the PICOLA processing unit 106 acquires audio data for the expansion processing data period Tg following the pitch W among the audio data held in the speech speed conversion processing buffer 102. . Then, the PICOLA processing unit 106 supplies the output buffer 107 with the audio data in which the generated audio data is inserted between the acquired audio data of the pitch W and the audio data of the decompression data period Tg (S213). ). If there is no audio data for the decompression processing data period Tg remaining in the speech speed conversion processing buffer 102, the PICOLA processing unit 106 determines the length (unprocessed data period Tu) of the shortage. Report to the speed control unit 104.

  When the processing in step S211 or S214 described above is completed, the pitch extraction unit 105 checks whether or not the speech data for the pitch search range Tp is held in the speech speed conversion processing buffer 102 (S214). .

  If the pitch extraction unit 105 confirms that the speech data for the pitch search range Tp is held in the speech speed conversion processing buffer 102 in step S214, the speech speed conversion device 100 determines that the speech speed conversion device 100 in step S209 described above. From the processing, the pitch candidate extraction processing is performed again.

  On the other hand, when it is confirmed that the speech speed conversion processing buffer 102 holds the speech data in the speech speed conversion processing buffer 102 for less than the pitch search range Tp, the speech speed conversion apparatus 100 performs the speech speed conversion. As long as the process continues, the operation returns to step S201 described above.

  Next, regarding the transition of voice data held in the speech speed conversion processing buffer 102 and the output buffer 107 when the speech speed conversion apparatus 100 operates according to the flowcharts of FIGS. 3 to 5 described above, FIGS. This will be described using the timing chart.

  Here, the description will be made assuming that the number of samples constituting the output frame and the input frame is n = 80, the expansion rate Rs = 1.25, and the pitch search range Tp = 240 samples (a period of Ts × 120).

  In the timing charts of FIGS. 6 to 9, the speech speed conversion apparatus 100 starts the speech speed conversion process from the time point T0. 6 to 9, timings T1, T2, T3, T4, T5... Indicate timings at which output frames should be output from the output buffer 107. For example, the time point of the timing T1 indicates a time point when 80 samples of time (Ts × 80 time) has elapsed from the timing T0, and the time point of the timing T2 indicates a time corresponding to 160 samples from the timing T0 (Ts × 160). This indicates the point in time when

  Here, for ease of explanation, first, transitions of speech data held in the speech speed conversion processing buffer 102 and the output buffer 107 when the minimum delay amount Td = pitch search range Tp are shown in FIGS. 7 for explanation.

  FIG. 6 shows transition of audio data held in the speech rate conversion processing buffer 102 and the output buffer 107 when the minimum delay amount Td = pitch search range Tp and the expansion processing by the PICOLA processing unit 106 is performed.

In Figure 6, at the point of time T3, since the minimum delay amount Td worth of audio data (240 samples) so that the accumulated, it is assumed that the pitch W ₁ of the length of Tw ₁ is extracted by the pitch extraction unit 105 . It is assumed that the pitch W ₁ is determined to be appropriate for application to the decompression process by the pitch extraction unit 105, and further, voice data of a speech waveform using the pitch W ₁ is generated by the PICOLA processing unit 106. Here, it is assumed that Tw ₁ = 45 samples. Then, in the speech speed control unit 104, the data period for expansion processing Tg ₁ = (Tw ₁ / (Rs−1)) − Tw ₁ = (45 / (1.25−1)) − 45 = period corresponding to 135 samples. It becomes.

Therefore, as shown in FIG. 6, the PICOLA processing unit 106, at timing T3, the audio data having the pitch W ₁ (Tw ₁ ), the audio data generated based on the pitch W ₁ , and the decompression processing data period Tg Audio data (180 samples) is held and supplied to the data output unit 108. That is, at this time, the data output unit 108 holds the sound data for 45 + 45 + 135 = 225 samples. At the time T3, the speech speed conversion processing buffer 102 holds only 240-225 = 15 samples of audio data. Therefore, at least until the time T6, the speech speed conversion processing buffer 102 Since no audio data (240 samples) corresponding to the pitch search range Tp is collected, the decompression process is not performed. Accordingly, new audio data is not supplied to the data output unit 108 during the period from the timing T3 to T6.

  Then, the output buffer 107 outputs the output frame F1 (80 samples of audio data) at the timing T3, and further outputs the output frame F2 (80 samples of audio data) at the timing T4. It will be. That is, since only 225-80-80 = 65 frames of audio data remain in the data output unit 108 at the timing T4, a normal output frame is generated and output even at the timing T5. You can't do that.

  As described above, when the minimum delay amount Td = the pitch search range Tp, after the first decompression process, before the data corresponding to the minimum delay amount Td necessary for the second pitch search is accumulated, the output buffer 107. In some cases, there is a situation in which data for one output frame is lost.

  In FIG. 7, the minimum delay amount Td = pitch search range Tp, and the speech speed conversion when the PICOLA processing unit 106 does not perform the expansion process (when the extracted pitch candidate is inappropriate for use in the expansion process). The transition of audio data held in the processing buffer 102 and the output buffer 107 is shown.

In FIG. 7, since the voice data (240 samples) corresponding to the minimum delay amount Td is accumulated at the timing T3, the pitch extraction unit 105 extracts the pitch W ₁ (pitch candidate) having the length of Tw _1. Shall be. Then, the pitch W ₁ is determined to be inappropriate by the pitch extraction unit 105 to be applied to the decompression process, and is output from the speech speed conversion processing buffer 102 to the output buffer 107. Here, it is assumed that Tw ₁ = 45 samples. Accordingly, in FIG. 7, the audio data for 45 samples is held in the output buffer 107 at the timing T3. However, since the audio data for one output frame (80 samples) is not reached, the output buffer 107 is normal. The state in which the output frame into which the audio data is inserted cannot be output continues.

  Next, as in the above-described embodiment, the speech speed when the pitch search range Tp = 240 and the minimum delay amount Td = 240 + 80-1 = 319 samples (the length calculated using the above equation (1)). Transition of audio data held in the conversion processing buffer 102 and the output buffer 107 will be described with reference to FIGS.

  In FIG. 8, the pitch search range Tp = 240 samples and the minimum delay amount Td = 319 samples, and the speech data stored in the speech speed conversion processing buffer 102 and the output buffer 107 when the expansion processing by the PICOLA processing unit 106 is performed are shown. Shows the transition. That is, the timing chart of FIG. 8 shows a case where the conditions are the same as those of FIG. 6 except for the minimum delay amount Td.

8, at the point of time T4, since the minimum delay amount Td min or more audio data (320 samples) would accumulate, and that the pitch W ₁ of the length of Tw ₁ is extracted by the pitch extraction unit 105 To do. It is assumed that the pitch W ₁ is determined to be appropriate for application to the decompression process by the pitch extraction unit 105, and further, voice data of a speech waveform using the pitch W ₁ is generated by the PICOLA processing unit 106. Here, it is assumed that Tw ₁ = 45 samples. Then, the speech speed control unit 104 calculates the decompression processing data period Tg ₁ = 135 samples, as in the case of FIG.

Further, as shown in FIG. 8, the PICOLA processing unit 106, at the timing T4, the audio data of the pitch W ₁ (Tw ₁ ), the audio data generated based on the pitch W ₁ , and the decompression processing data period Tg The audio data (135 samples) is held and supplied to the data output unit 108. That is, at this time, the data output unit 108 holds the audio data for 225 samples. At timing T4, the speech speed conversion processing buffer 102 holds 320-225 = 95 samples of audio data. Then, in the speech speed conversion processing buffer 102, voice data of the pitch search range Tp or more accumulates in the speech speed conversion processing buffer 102 at timing T6, and new voice data is supplied to the data output unit 108. Become.

  On the other hand, the output buffer 107 outputs an output frame F1 (80 samples of audio data) at timing T4, and further outputs an output frame F2 (80 samples of audio data) at timing T5. It will be. In the example of FIG. 8, new audio data is supplied to the output buffer 107 at the timing T6.

  Therefore, in the example of FIG. 8, by setting the minimum delay amount Td as described in (1) above, unlike the example of FIG. 6, the timing at which the output buffer 107 should output the output frame has arrived. In addition, the output buffer 107 can hold data for one output frame or more.

  In FIG. 9, the audio data held in the speech speed conversion processing buffer 102 and the output buffer 107 when the pitch search range Tp = 240 samples and the minimum delay amount Td = 319 samples and the expansion processing by the PICOLA processing unit 106 is not performed. It shows about the transition. That is, the timing chart of FIG. 9 shows a case where the conditions are the same as those of FIG. 7 except for the minimum delay amount Td.

In FIG. 9, since voice data (320 samples) equal to or greater than the minimum delay amount Td is accumulated in the speech speed conversion processing buffer 102 at the timing T4, the pitch extraction unit 105 causes the pitch of the length of Tw ₁ to be accumulated. It is assumed that W ₁ (pitch candidate) has been extracted. Then, the pitch W ₁ is determined to be inappropriate by the pitch extraction unit 105 to be applied to the decompression process, and is output from the speech speed conversion processing buffer 102 to the output buffer 107. Here, it is assumed that Tw ₁ = 45 samples.

At this time, since the voice data of 320−45 = 275 samples remains in the speech speed conversion processing buffer 102 and is larger than the pitch search range Tp (240 samples), the pitch extraction unit 105 performs timing T4. At this point, it is possible to continuously search for pitch candidates. Then, it is assumed that the pitch extraction unit 105 extracts a pitch W ₂ (pitch candidate) having a length of Tw _{2 in} the _second pitch candidate search. Then, it is assumed that the pitch W ₂ is determined to be inappropriate for use in the decompression process by the pitch extraction unit 105 and is output from the speech speed conversion processing buffer 102 to the output buffer 107. Here, it is assumed that Tw ₂ = 45 samples.

  Then, the data output unit 108 can hold 45 + 45 = 90 samples of audio data at the timing T4, and can output an output frame unlike the example of FIG.

  Therefore, in the example of FIG. 9, unlike the example of FIG. 7, even if the pitch candidates are supplied to the output buffer 107 as they are, voice data of the pitch search range Tp or more remains in the speech speed conversion processing buffer 102. The pitch extraction unit 105 can perform a pitch search a plurality of times within the same frame (same timing). That is, in the example of FIG. 9, by setting the minimum delay amount Td as in the above equation (1), unlike the example of FIG. 7, the timing at which the output buffer 107 should output an output frame has arrived. However, there is a high possibility that the output buffer 107 can hold data for one output frame or more.

(A-3) Effects of First Embodiment According to the first embodiment, the following effects can be achieved.

(A-3-1) In the speech speed conversion apparatus 100, the speech data of the pitch search range Tp or more is initially held in the speech speed conversion processing buffer 102 before the expansion process by the PICOLA processing unit 106 is started. The output buffer 107 can easily hold audio data for one output frame or more.

(A-3-2) The amount of data held in the speech speed conversion processing buffer 102 in the initial stage can make the data amount held in the output buffer 107 less likely to be exhausted. If the time from when audio data is input to the data input unit 101 to when it is output by the data output unit 108 increases, the processing quality (real-time performance) deteriorates. Therefore, in the speech speed conversion apparatus 100 of this embodiment, “pitch search range Tp + 1 frame−1 sample is used as the minimum minimum delay amount Td necessary for holding the audio data for one output frame or more in the output buffer 107. Is set.

  Next, the reason why “pitch search range Tp + 1 frame−1 sample” is desirable as the minimum delay amount Td will be described.

  As an example of the problem that data for one output frame disappears in the output buffer 107, the above-described condition of FIG. 6 (hereinafter referred to as “first condition”) and the above-described condition of FIG. (Referred to as “second condition”). Of these, the second condition can be regarded as the same as when the expansion rate of the first example is 1.0. That is, if it can be shown that the problem that occurs under the second condition can be avoided, the problem that occurs under the first condition can also be avoided. This is because the decompression process is always performed under the decompression rate Rs> 1.0 in the first condition, so that the data in the output buffer 107 increases as compared with the second condition.

  Even in the second condition, if the period for the pitch candidate W (hereinafter referred to as “TW”) is the same as the period for one output frame (hereinafter referred to as “TF”) or a multiple of TF, The problem does not occur. This is because the period during which the output buffer 107 is supplied is always a multiple of the period corresponding to TF, so that the input data can be supplied as it is. The above-described problem occurs when the period TW of the pitch candidate W has a value other than that. Here, for each “period (data length)”, the number of samples constituting the audio data (all sampling intervals are assumed to be the same) is represented as a unit.

  The following shows how a desirable value is obtained as the minimum delay amount Td that can avoid the above-described problem in the above-described second example (example shown in FIG. 7). In the following example, the maximum value of TW is Tp / 2, but the present invention is not limited to this, and extraction may be performed up to Td. In the following example, Tp> = TF. In the following example, it is assumed that TF = 80 (sample) and Tp = 240 (sample).

  First, the case where TW is less than 1 * TF (when 1 <= TW <= 79) will be described.

  In this case, in the first pitch search (pitch extraction) shown in FIG. 7, since data for TF is accumulated in the output buffer 107, it is necessary to perform pitch extraction once again within the same frame processing. Assuming that the period of the voice data held in the speech speed conversion processing buffer 102 is P and the initial value is X, the period after the first pitch extraction is (X-1) <= P <= (X-79). is there. At this time, since P may be data equal to or more than Tp, the required maximum value of X is X−79 <= 24 and X = 319.

  Next, the case where TW is larger than 1 * TF and smaller than 2 * TF (when 81 <= TW <= 159) will be described.

  In this case, as a result of the first pitch extraction, the period held in the speech speed conversion processing buffer 102 is (X−81) <= P <= (X−159). Further, since the output buffer 107 exists in the range of 1 * TF or more and less than 2 * TF, output frames from the output buffer 107 can be output. However, in the output buffer 107, when the next output frame is to be output, the data for TF does not remain. Therefore, it is necessary to always perform pitch extraction at the output timing of the next output frame. is there. Therefore, in this case, since it is sufficient that data equal to or more than Tp exists in the speech rate conversion processing buffer 102 after the data for DF is input at the timing at which the next output frame is to be output, the required maximum value of X is X−. 159 + 80 = 240 and X = 319. The value of X is the minimum delay amount Td, which is the same when 2 * F <W <3 * F, 3 * F <W <4 * F.

  From the above, the minimum Td at which the above-described problem is difficult to occur can be derived from the following equation (4). In the following equation (4), it is assumed that TW ≠ n * TF (n = 0, 1, 2,...) And Tp> TF. In the following expression (4), “int (TW / TF) indicates an integer part of“ TW / TF ”. Further, in the following expression (4), “max {TW−int (W / TF) * TF}” is “TW−int” when n is changed to “n = 0, 1, 2,... The maximum value of (W / TF) * TF ”is shown, and when calculated, the result shown in the following equation (5) is obtained. Therefore, as shown in the following equation (6), a result that the above-described “pitch search range Tp + 1 frame−1 sample” is desirable as Td is obtained.

Td = Tp + max {TW−int (TW / TF) * TF} (4)
max {TW-int (TW / TF) * TF} = TF-1 (5)
Td = Tp + TF-1 (6)
As described above, in the speech rate conversion apparatus 100, by storing audio data for one output frame or more in the output buffer 107, it is possible to execute the speech rate conversion process with stable accuracy in real time.

  As a result, by using the speech rate conversion device 100, it is possible to perform speech rate conversion on devices that process voice in real time, such as televisions, radios, and telephones. Further, by using the speech speed conversion device 100, the user can perform the speech speed conversion at any time during use of these devices, and can change the speech speed to an arbitrary speech speed even during the execution of the speech speed conversion. be able to.

(B) Second Embodiment Hereinafter, a second embodiment of an audio signal processing apparatus and program according to the present invention will be described in detail with reference to the drawings. Note that the audio signal processing device according to the first embodiment is a speech rate conversion device.

(B-1) Configuration of Second Embodiment FIG. 10 is a block diagram showing a functional configuration of a speech rate conversion apparatus 100A of the second embodiment. In FIG. 10, the same or corresponding reference numerals are given to the same or corresponding parts as those in FIG. Hereinafter, the difference between the second embodiment and the first embodiment will be described.

  The speech speed conversion apparatus 100A is different from the first embodiment in that a speech section detection unit 109 and a delay recovery unit 110 are added.

  In the speech speed conversion apparatus 100 according to the first embodiment, if the speech speed conversion process is continued, the speech signal (speech data) is expanded one after another, so that the delay increases indefinitely and is held in the output buffer 107 accordingly. The voice data to be increased also increases. As a result, for example, when the speech speed conversion device 100 is applied to telephone communication such as a telephone device and the speech speed conversion processing is performed in real time, a problem occurs in communication between speakers.

  Therefore, in the speech speed conversion device 100A of the second embodiment, the speech section detection unit 109 and the delay recovery unit 110 are added to correspond to the delay recovery function, and cope with the above-described problems.

Note that the delay here refers to, for example, the time corresponding to the latest voice data input in the speech speed conversion processing buffer 102 by the decompression process in the speech speed conversion device (the time on the sample time series), This is the difference from the time corresponding to the audio data output from the output buffer 107.

  The delay recovery function is, for example, a speech speed conversion device that, when a delay of a certain time or more has occurred due to decompression processing, a non-speech segment (silence segment) included in the latest input speech signal. This is a function for shortening the delay time by deleting from the fast conversion processing buffer 102.

  The voice section detection unit 109 determines whether the voice signal indicated by the voice data of the input frame is a voice section (sound section) with respect to one frame of voice data (voice signal) acquired by the data input unit 201. It is determined whether it is a voice section (silent section). An existing processing configuration (for example, a method using input signal power, zero-crossing, correlation function, etc. as an acoustic feature amount) can be applied to the processing of detecting a voiced section in the voice section detection unit 109. Note that the speech segment detection unit 109 may be incorporated in the data input unit 101 and the speech segment detection process may be performed when the data input unit 101 generates an input frame.

  Then, when the acquired input frame is a non-speech section, the speech section detection unit 109 delivers the input frame to the delay recovery unit 110.

  When the input frame of the non-speech section is supplied from the speech section detection unit 109, the delay recovery unit 110 controls the speech speed conversion processing buffer 203 to output all the accumulated speech data to the output buffer 208. Then, the delay recovery unit 110 deletes (discards) the most recently acquired input frame when the amount of audio data accumulated in the output buffer 107 is greater than or equal to a predetermined amount (in this example, the minimum delay amount is Td or more). And delay recovery. On the other hand, when the amount of audio data accumulated in the output buffer 107 is less than a predetermined value (here, it is assumed that it is less than the minimum delay amount Td as an example), the delay recovery unit 110 converts the latest acquired input frame into the output buffer 107. To supply.

(B-2) Operation | movement of 2nd Embodiment Next, operation | movement of the speech-speed converter 100A of 2nd Embodiment which has the above structures is demonstrated.

  The speech speed conversion apparatus 100A according to the second embodiment is the same as the first embodiment except that the operations of the speech section detection unit 109 and the delay recovery unit 110 related to the above-described delay recovery function are added. . Specifically, in the second embodiment, the operations of the speech section detection unit 109 and the delay recovery unit 110 related to the delay recovery function described above are performed between step S201 and step S202 of the operation of the first embodiment. Inserted.

  In FIG. 11, in the speech speed conversion apparatus 100A of the second embodiment, the operations of the speech section detection unit 109 and the delay recovery unit 110 related to the delay recovery function inserted between the above-described steps S201 and S202. Show. Although not shown in FIG. 11, other operations are the same as those in the first embodiment.

  In the speech rate conversion apparatus 100A of the second embodiment, when an input frame is acquired by the data input unit 101 in step S201 described above, the speech segment detection unit 109 determines whether or not the input frame is a speech segment. Is determined (S301).

  When it is determined in step S301 that the input frame is a speech section, the speech speed conversion apparatus 100A inputs the input frame to the speech speed conversion processing buffer 102 in step S202 described above, and the subsequent processing is the first process. The processing is the same as that of the embodiment.

  On the other hand, when it is determined in step S301 that the input frame is a speech section, the speech section detection unit 109 supplies the input frame to the delay recovery unit 110. Then, the delay recovery unit 110 controls the speech speed conversion processing buffer 203 to output all accumulated audio data to the output buffer 208 (S302).

  Then, the delay recovery unit 110 determines whether the amount of audio data accumulated in the output buffer 107 is equal to or greater than the minimum delay amount Td (S303).

  If it is determined in step S303 described above that the amount of audio data accumulated in the output buffer 107 is equal to or greater than the minimum delay amount Td, the delay recovery unit 110 deletes (discards) the most recently acquired input frame. Then, delay recovery is attempted (S304). Then, the speech speed conversion apparatus 100A returns to the above-described step S201 and operates. That is, the delay recovery unit 110 increases the delay amount to a predetermined amount or more when the total amount of audio data accumulated in the speech speed conversion processing buffer 102 and the output buffer 107 is equal to or greater than the minimum delay amount Td. And delay recovery processing (discarding input frames in non-voice segments) is performed.

  On the other hand, if it is determined in step S303 described above that the amount of audio data accumulated in the output buffer 107 is less than the minimum delay amount Td, the delay recovery unit 110 outputs the most recently acquired input frame as an output. The data is supplied to the buffer 107 (S305). Then, the speech speed conversion apparatus 100A returns to the above-described step S201 and operates.

(B-3) Effects of the Second Embodiment According to the second embodiment, the following effects can be obtained in addition to the first embodiment.

  In the speech rate conversion apparatus 100A, the data of the minimum delay amount Td is held in the output buffer 107, and the voice data in the non-voice section is deleted (discarded). Thereby, in the speech speed conversion apparatus 100A, the delay can be shortened (delay recovery) without causing a situation in which the audio data for one output frame is not in the output buffer 107. As a result, the speech speed conversion apparatus 100A can perform stable speech speed conversion processing while suppressing the amount of memory required for the output buffer 107 and suppressing the delay amount.

(C) Other Embodiments The present invention is not limited to the above-described embodiments, and may include modified embodiments as exemplified below.

(C-1) In the speech speed conversion device of each embodiment described above, only the case of performing the process of reducing the speech speed has been described. However, in the speech speed conversion device of each of the above embodiments, the process of increasing the speech speed. (For example, an existing speech speed conversion process may be applied).

(C-2) In the speech speed conversion device of each of the embodiments described above, the delay control unit accumulates voice data of the minimum delay amount Td in the speech speed conversion processing buffer only immediately after the speech speed conversion device starts processing. Up to this point, the speech speed conversion process by the PICOLA processing unit is suspended, but the same process may be performed at other timings.

  For example, in the speech speed conversion apparatus, when the delay amount becomes a certain amount or more, the speech data held in the speech speed conversion processing buffer 102 and the output buffer 107 is refreshed (all data is deleted), and then the speech speed is changed. The speech speed conversion process by the PICOLA processing unit may be suspended until voice data having the minimum delay amount Td is accumulated in the conversion processing buffer.

(C-3) In each of the embodiments described above, the delay control unit stops the expansion process when the amount of delay exceeds a certain amount in the speech speed conversion device in order to prevent the delay from increasing indefinitely due to the expansion process. As described above, it may be made to correspond to a function (delay limiting function) for controlling another processing configuration.

  DESCRIPTION OF SYMBOLS 100 ... Speech speed converter 101 ... Data input part 102 ... Speech speed conversion process buffer 103 ... Delay control part 104 ... Speech speed control part 105 ... Pitch extraction part 106 ... PICOLA process part 107 ... Output buffer 108: Data output unit.

Claims (4)

Input buffer means for storing audio data of the input audio signal;
For the audio signal waveform based on the audio data stored in the input buffer means, the basic period is extracted from the audio signal waveform for the search period, and the extracted audio signal waveform of the basic period is used for the input buffer means. Speech rate conversion means for performing speech rate conversion processing on the accumulated voice data;
Output buffer means for storing voice data after the speech speed conversion means performs the speech speed conversion processing;
Audio data output means for outputting an output audio data frame including audio data for the output interval among audio data stored in the output buffer means for each output interval;
The speech rate conversion is performed after the search period and the output interval are added to the input buffer means, and further, the voice data of the minimum accumulation period obtained by subtracting one sampling period of the voice data of the output buffer means is accumulated. An audio signal processing apparatus comprising: conversion processing control means for starting speech speed conversion processing by the means. In the audio signal processing device, when the audio data input to the input buffer means is audio data in a non-audio section in a state where the delay amount associated with the speech speed conversion process is greater than or equal to a predetermined value, The audio signal processing apparatus according to claim 1, further comprising delay recovery means for discarding data and performing delay recovery. The delayed recovery means, claim, characterized in that it is determined that the amount of audio data stored in said input buffer means and said output buffer means when a predetermined or more, the delay amount becomes a predetermined value or more 2 The audio signal processing apparatus according to 1. Computer
Input buffer means for storing audio data of the input audio signal;
For the audio signal waveform based on the audio data stored in the input buffer means, the basic period is extracted from the audio signal waveform for the search period, and the extracted audio signal waveform of the basic period is used for the input buffer means. Speech rate conversion means for performing speech rate conversion processing on the accumulated voice data;
Output buffer means for storing voice data after the speech speed conversion means performs the speech speed conversion processing;
Audio data output means for outputting an output audio data frame including audio data for the output interval among audio data stored in the output buffer means for each output interval;
After the search buffer and the output interval are added to the input buffer means, and further, the audio data of the minimum accumulation period that is less than the minimum accumulation period obtained by subtracting one sampling period of the audio data of the output buffer means is accumulated. An audio signal processing program that functions as conversion processing control means for starting speech speed conversion processing by the speech speed conversion means.

Images