JP4461985B2 - Speech waveform expansion device, waveform expansion method, speech waveform reduction device, waveform reduction method, program, and speech processing device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To realize speech reduction and expansion with arbitrary magnification while reducing deterioration of a speech waveform. <P>SOLUTION: A speech processor 100 divides an inputted speech signal into pitch waveforms. Similarities of each pitch waveform with pitch waveforms right before and after it are calculated and pitch waveforms are generated together with pitch waveforms with higher similarities. The generated pitch waveforms are inserted between the source pitch waveforms for the generation when expanded or replaced with the source pitch waveforms for the generation when reduced. The pitch waveforms are processed in the decreasing order of similarities within processing units. <P>COPYRIGHT: (C)2006,JPO&amp;NCIPI

Description

  The present invention relates to an audio waveform expansion device, a waveform expansion method, an audio waveform reduction device, a waveform reduction method, a program, and an audio processing device, and in particular, an audio waveform expansion device that extends and outputs an input audio waveform in time, Waveform decompression method and program, speech waveform reduction device for reducing and outputting input speech waveform in time, waveform reduction method and program, and speech processing for extending and outputting input speech waveform in time Relates to the device.

  As one of the processes for transforming audio data, there is a TDHS (Time Domain Harmonic Scaling) system that reduces and expands the length of the audio waveform to m / n times (m and n are natural numbers) in time (for example, patents). Reference 1). FIG. 7 is a diagram for explaining the principle of the TDHS method.

  As shown in the figure, in the TDHS system, a partial speech waveform having a length mT starting from a location 0 where processing is started this time (T is a length of one cycle of a repeated waveform) and a current location (nm). ) Weighted addition of the partial speech waveform of length mT starting from the location of T, and replaces the portion from 0 to nT with the generated partial speech waveform (length mT). That is, the length is increased or decreased by (mn) T by one replacement of the partial speech waveform. By repeating this, the length of the speech waveform as a whole is reduced / expanded to m / n times (m and n are natural numbers) in terms of time.

In this case, of the two partial speech waveforms to be weighted and added, the weight W (k) is used for the partial speech waveform on the past side in time, and the weight 1 is used for the partial speech waveform on the future side in terms of time. -W (k) is multiplied. Here, the value of W (k) changes linearly from the value 0 to the value 1 from the head sample position to the tail sample position of the partial speech waveform. By using such weighting factors W (k) and 1-W (k), the waveform can be reduced / expanded while maintaining continuity.
JP-A-8-146993 (page 3-5, FIG. 12-15)

  In the TDHS system, the signal waveform can be reduced / expanded only by an arbitrary rational multiple (m / n).

  The present invention has been made in view of the above problems, and is a speech waveform expansion device, a waveform expansion processing method and a program that can expand a signal waveform at an arbitrary magnification, and reduce a signal waveform at an arbitrary magnification. It is an object of the present invention to provide a voice waveform reduction device, a waveform reduction processing method and program, and a voice processing device that can expand or reduce a signal waveform at an arbitrary magnification.

A speech waveform expansion device according to a first aspect of the present invention is:
A speech waveform expansion device for expanding and outputting an input waveform on a time axis,
Input waveform receiving means for receiving data representing the input waveform;
An expansion ratio acceptance means for accepting an input of a magnification for expanding the input waveform;
Cutting means for cutting out a pitch waveform from the input waveform received by the input waveform receiving means;
For each pitch waveform cut out by the cut-out means, similarity calculation means for calculating the similarity between the pitch waveform and one pitch waveform adjacent to the pitch waveform before and after on the time axis; ,
Ordering means for assigning a processing order to each pitch waveform using the similarity calculated by the similarity calculation means as a scale;
Expanded waveform generation that generates a waveform for insertion by selecting a pitch waveform in the processing order given by the ordering means and weighting and adding the selected pitch waveform and one pitch waveform adjacent to the pitch waveform Means,
An expanded waveform connecting means for inserting the waveform generated by the expanded waveform generating means between two pitch waveforms that are subjected to weighted addition on the input waveform;
Equipped with,
Repeating the processing by the expanded waveform generating means and the expanded waveform connecting means while updating the pitch waveform to be selected according to the processing order until the waveform length of the specified magnification is reached.
It is characterized by.

  According to the present invention, a stretched waveform is generated by selecting a pitch waveform in order from a waveform having a high degree of similarity, that is, a portion where similar waveforms are repeated. For this reason, it is possible to expand and output the input sound to an arbitrarily specified magnification while suppressing deterioration of sound quality.

In the speech waveform expansion device,
The cutout means cuts out a partial waveform having a certain length from the input waveform, and cuts out a pitch waveform in units of cutout partial waveforms.
In this case, the ordering means performs processing of ordering the pitch waveforms in units of the fixed-length partial waveform.

  According to the present invention, when the pitch changes midway, the input waveform can be reduced / expanded following the change.

The waveform expansion method according to the second aspect of the present invention is:
An input waveform receiving step for receiving data representing the input waveform;
An expansion magnification reception step for receiving an input of a magnification for expanding the input waveform;
A step of cutting out a pitch waveform from the input waveform received in the input waveform receiving step;
For each pitch waveform cut out in the cut-out step, a similarity calculation step for calculating a similarity between the pitch waveform and one pitch waveform adjacent to the pitch waveform before and after on the time axis; ,
An ordering step of assigning a processing order to each pitch waveform using the similarity calculated in the similarity calculation step as a scale;
Choose Back Symbol ordered pitch waveforms applied treatment sequence in step, by weighted addition of the one pitch waveform adjacent to the selected pitch waveform and the pitch waveform, decompression waveform to generate a waveform for insertion Generation step;
An expanded waveform connecting step for inserting the waveform generated in the expanded waveform generating step between two pitch waveforms that are subjected to weighted addition on the input waveform;
With
Until waveform length of the specified magnification, while updating the pitch waveforms to be selected according to the processing order, and repeating said expansion waveform generation step, a process by said expansion waveform connecting step.

The program according to the third aspect of the present invention is:
The computer used in the speech waveform decompression apparatus for decompressing and outputting input waveform on the time axis,
Input waveform receiving means for receiving data representing the input waveform ;
And stretching magnification accepting means for accepting an input of a magnification stretching the input waveform,
And the cut-out means that you cut out the pitch waveform from the input waveform received by the input waveform receiving means,
For each pitch waveform cut out by the cut- out means, similarity calculation means for calculating the similarity between the pitch waveform and one pitch waveform adjacent to the pitch waveform before and after on the time axis; ,
Ordering means for assigning a processing order to each pitch waveform using the similarity calculated by the similarity calculation means as a scale;
Expanded waveform generation that generates a waveform for insertion by selecting a pitch waveform in the processing order given by the ordering means and weighting and adding the selected pitch waveform and one pitch waveform adjacent to the pitch waveform Means ,
An expanded waveform connecting means for inserting the waveform generated by the expanded waveform generating means between two pitch waveforms that are subjected to weighted addition on the input waveform;
To function,
Until waveform length of the specified magnification, while updating the pitch waveforms to be selected according to the processing order, and said decompressed waveform generating means, that it has the decompressed waveform connecting means and repeatedly to by the processing returns Suyo Features .

The speech waveform reduction device according to the fourth aspect of the present invention is:
A speech waveform reduction device that reduces and outputs an input waveform on a time axis,
Input waveform receiving means for receiving data representing the input waveform;
A reduction magnification receiving means for receiving an input of a magnification for reducing the input waveform;
Cutting means for cutting out a pitch waveform from the input waveform received by the input waveform receiving means;
For each pitch waveform cut out by the cut-out means, similarity calculation means for calculating the similarity between the pitch waveform and one pitch waveform adjacent to the pitch waveform before and after on the time axis; ,
Ordering means for assigning a processing order to each pitch waveform using the similarity calculated by the similarity calculation means as a scale;
Reduced waveform generation that generates a replacement waveform by selecting a pitch waveform in the processing order given by the ordering means and weighting and adding the selected pitch waveform and one pitch waveform adjacent to the pitch waveform Means,
The waveform generated by the reduced waveform generating means, and two reduced waveform connecting means Ru replaced with pitch waveform subject to weighting addition in the reduced waveform generating means on the input waveform,
Equipped with,
Repeating the processing by the reduced waveform generating means and the reduced waveform connecting means while updating the pitch waveform to be selected according to the processing order until the waveform length of the specified magnification is reached.
It is characterized by.

  According to the present invention, a reduced waveform is generated by selecting a pitch waveform in order from the highest waveform similarity, that is, from a portion where similar waveforms are repeated. For this reason, it is possible to reduce the input voice to an arbitrarily specified magnification and output the output while suppressing deterioration in sound quality.

In the speech waveform reduction device,
The cutout means cuts out a partial waveform having a certain length from the input waveform, and cuts out a pitch waveform in units of cutout partial waveforms.
In this case, the ordering means performs processing of ordering the pitch waveforms in units of the fixed-length partial waveform.

The waveform reduction method according to the fifth aspect of the present invention is:
An input waveform reception step for receiving data representing the input waveform;
A reduction magnification acceptance step for accepting an input of a magnification for reducing the input waveform;
A step of cutting out a pitch waveform from the input waveform received in the input waveform receiving step;
For each pitch waveform cut out in the cut-out step, a similarity calculation step for calculating a similarity between the pitch waveform and one pitch waveform adjacent to the pitch waveform before and after on the time axis; ,
An ordering step of assigning a processing order to each pitch waveform using the similarity calculated in the similarity calculation step as a scale;
Choose Back Symbol ordered pitch waveform processing order granted in step, by weighted addition of the one pitch waveform adjacent to the selected pitch waveform and the pitch waveform, reduced to generate a waveform for replacement waveform Generation step;
The waveform generated by the reduced waveform generation step, and two reduced waveform connecting step of Ru replaced with pitch waveform subject to weighting addition in the reduced waveform generation step on the input waveform,
With
Until waveform length of the specified magnification, while updating the pitch waveforms to be selected according to the processing order, and repeating said reduced waveform generation step, a process by said reduction waveform connecting step.

The program according to the sixth aspect of the present invention is:
The computer used to input waveform to the sound wave reduction device for outputting reduced on the time axis,
Input waveform receiving means for receiving data representing the input waveform ;
A reduction ratio accepting means for accepting an input of a magnification reducing the input waveform,
And the cut-out means that you cut out the pitch waveform from the input waveform received by the input waveform receiving means,
For each pitch waveform cut out by the cut- out means, similarity calculation means for calculating the similarity between the pitch waveform and one pitch waveform adjacent to the pitch waveform before and after on the time axis; ,
Ordering means for assigning a processing order to each pitch waveform using the similarity calculated by the similarity calculation means as a scale;
Reduced waveform generation that generates a replacement waveform by selecting a pitch waveform in the processing order given by the ordering means and weighting and adding the selected pitch waveform and one pitch waveform adjacent to the pitch waveform Means ,
The waveform generated by the reduced waveform generating means, a reduced waveform connecting means for replacing two pitch waveforms subject to weighting addition in the reduced waveform generating means on the input waveform,
To function,
Until waveform length of the specified magnification, while updating the pitch waveforms to be selected according to the processing order, and the reduced waveform generating means, that it has to the reduced waveform connecting means and repeatedly to by the processing returns Suyo Features .

The speech processing apparatus according to the seventh aspect of the present invention is:
A speech processing apparatus that outputs an input waveform by extending or reducing on a time axis,
Input waveform receiving means for receiving data representing the input waveform;
Magnification accepting means for accepting an input of a magnification for expanding or reducing the input waveform;
Cutting means for cutting out a pitch waveform from the input waveform received by the input waveform receiving means;
For each pitch waveform cut out by the cut-out means, similarity calculation means for calculating the similarity between the pitch waveform and one pitch waveform adjacent to the pitch waveform before and after on the time axis; ,
Ordering means for assigning a processing order to each pitch waveform using the similarity calculated by the similarity calculation means as a scale;
Expanded waveform generation that generates a waveform for insertion by selecting a pitch waveform in the processing order given by the ordering means and weighting and adding the selected pitch waveform and one pitch waveform adjacent to the pitch waveform Means,
An expanded waveform connecting means for inserting the waveform generated by the expanded waveform generating means between two pitch waveforms that are subjected to weighted addition on the input waveform;
Reduced waveform generation that generates a replacement waveform by selecting a pitch waveform in the processing order given by the ordering means and weighting and adding the selected pitch waveform and one pitch waveform adjacent to the pitch waveform Means,
The waveform generated by the reduced waveform generating means, and two reduced waveform connecting means Ru replaced with pitch waveform subject to weighting addition in the reduced waveform generating means on the input waveform,
Magnification discrimination means for discriminating whether to expand or reduce the input waveform;
When it is determined that the input waveform is to be expanded by the magnification determining means, the expanded waveform generating means is updated while updating the pitch waveform selected according to the processing order until the waveform length of the specified magnification is reached. And first repeating means for repeating the processing by the expanded waveform connecting means,
When it is determined that the input waveform is to be reduced by the magnification determining means, the reduced waveform generating means is updated while updating the pitch waveform selected according to the processing order until the waveform length of the specified magnification is reached. And second repeating means for repeating the processing by the reduced waveform connecting means,
It comprises.

  According to the present invention, it is possible to reduce or expand a speech waveform at an arbitrary magnification while suppressing deterioration in sound quality.

  Embodiments according to the present invention will be described below with reference to the drawings.

(Embodiment 1)
In the first embodiment, the case where the specified magnification is between 1/2 and 2 will be described first, and the case where the specified magnification is between 0 and 1/2 or 2 or more will be described later. If the specified magnification is greater than 1, the input audio signal waveform is expanded and output. If it is less than 1, the input audio signal waveform is reduced and output.

FIG. 1 is a block diagram showing a configuration of a sound processing apparatus according to an embodiment of the present invention. As shown in FIG. 1, the audio processing device 100 is configured by an information processing device such as a computer, for example. The input device 12, the output device 13, and the recording medium 17 are connected to the sound processing device 100. In response to the instruction from the input device 12, the audio processing device 100 expands or reduces the audio waveform data input from the recording medium 17 to the length of the designated magnification, and outputs it to the recording medium 17.

  Here, the audio waveform data is sample value data in which analog audio is quantized at a predetermined sampling frequency (for example, 8 kHz).

  The recording medium 17 is, for example, a CD-RW (Compact Disk ReWritable) disk and stores audio waveform data.

  The speech processing apparatus 100 includes a control unit 110, an input control unit 120, an output control unit 130, a program storage unit 140, a storage unit 150, and a data recording unit 170.

  The control unit 110 includes, for example, a central processing unit (CPU), a random access memory (RAM), and the like, and is based on an operation program stored in advance in the program storage unit 140. Control each unit, read out the audio waveform data stored in the recording medium 17 via the data recording unit 170, write the expanded / reduced audio waveform data into the recording medium 17, Perform waveform reduction processing.

  The control unit 110 performs waveform expansion processing or waveform reduction processing on the audio waveform data temporarily stored in the storage unit 150, and stores the expanded and reduced audio waveform data in the storage unit 150. In the case of the waveform expansion processing, the control unit 110 divides the audio waveform data into several parts (hereinafter referred to as pitch waveforms) in units of repetition, and based on each pitch waveform and one of the pitch waveforms before and after that. Then, a speech waveform is generated so as to have a specified magnification, and inserted between the pitch waveforms from which the generated speech waveform is generated. In the case of the waveform reduction processing, the control unit 110 divides the voice waveform data into pitch waveform units, and the voice waveform is set to a specified magnification based on each pitch waveform and one of the pitch waveforms before and after the pitch waveform. Is replaced with the speech waveform of the section having the pitch waveform from which the speech waveform is generated as the head and tail.

  When generating the pitch waveform, the control unit 110 discriminates the one having a higher correlation with the pitch waveform of interest among the pitch waveforms before and after the pitch waveform. High correlation means that the two pitch waveforms are similar. The more the pitch waveform is generated from the more similar pitch waveform, the more the deterioration of the obtained expansion (reduction) pitch waveform can be suppressed.

  The control unit 110 reads out and executes an operation program or the like stored in advance in the program storage unit 140, thereby, for example, as shown in FIG. 2, a division unit 111, a pitch extraction unit 112, a pitch selection unit 113, a waveform reduction / reduction unit, and the like. The decompression unit 114 and the like are realized.

  The dividing unit 111 divides the input audio signal into audio frames having a predetermined time length (number of samples M), and transmits the audio frames to the pitch extracting unit 112 and the pitch reducing / expanding unit. Note that the length of the voice frame needs to be sufficiently longer than the pitch waveform to be discriminated so that the pitch waveform can be discriminated / cut out within one voice frame. From experience, the length of the voice frame needs to be at least about 3.4 to 4 times the length of the pitch waveform included in the voice frame.

The pitch extraction unit 112 determines a pitch waveform present in the audio frame. For example, when each sample value in the voice frame is set as {s ₀ , s ₁ ,..., S _M−1 }, the calculation formula shown by the following equation 1 is calculated from t _start to t _end. Of these, _t that minimizes _et is the length of the pitch waveform (hereinafter referred to as N). Here, t _start and t _end can be changed within a reasonable range that the pitch length is within the range according to the audio signal to be reduced / expanded. For example, when human speech is to be reduced / expanded, t _start is the number of samples corresponding to about 400 Hz (sampling rate / 400), and t _end is the number of samples corresponding to about 50 Hz (sampling rate / 50). To do.

  As another example of the pitch extraction method, a method of cutting the pitch waveform by estimating the pitch length from the sample point where the sample value takes the minimum value or the maximum value, or from the sample point where the sign value of the sample value changes can be considered. . For example, if the audio frame of the sample point taking the minimum value is 6, 12, 35, 42, 66, 72, 95, 102, 126,. And the length of the pitch waveform is estimated to be 60.

  When the pitch waveform is cut out from the voice frame according to the length of the pitch waveform calculated by the pitch extraction unit 112, a portion less than the length of the pitch waveform remains at the end of the voice frame. Therefore, the pitch extraction unit 112 uses the remaining part in the next processing. Therefore, the pitch extraction unit 112 extracts a pitch waveform from the combination of the current unprocessed portion and the audio frame at the next processing. Of course, when the last audio frame is being processed, the remaining portion is output as it is.

Thus, the reason why the pitch extraction unit 112 extracts the pitch waveform in units of audio frames is to cope with the possibility that the pitch length of the input waveform changes in the middle.

The pitch selection unit 113 orders the pitch waveforms of the audio data divided into pitch waveform units by the pitch extraction unit 112 according to a predetermined standard. The predetermined standard is a standard that the sound quality is not easily deteriorated even if the pitch waveform is reduced / expanded. As a specific example, as shown in Formula 2, the order of the mean square error is ascending. Here, S _x is the sample value in the x-th position from the head of the audio frame, Pl _n is the number of samples n-th pitch waveform from the beginning, the beginning of a sample of p _n is the n-th pitch waveform from the head This is the number of values counted from the beginning in the audio frame.

  The waveform reduction / expansion unit 114 generates a new pitch waveform from the continuous pitch waveform, and replaces and inserts the pitch waveform in accordance with the reduction and extension. Here, the pitch waveform for generating a new pitch waveform is a focused pitch waveform and one pitch waveform before and after the pitch waveform. Which pitch waveform is selected from the pitch waveforms before and after the pitch waveform of interest is as follows.

That is, the sample value sequence of the pitch waveform of interest is {x _k , x _{k + 1} ,..., X _{k + N−1} }, and the sample value sequence of the pitch waveform before this pitch waveform is {x _k−N , X _{k−N + 1} ,..., X _k−1 }, and the sample value sequence of the pitch waveform after this pitch waveform is {x _{k + N} , x _{k + N + 1} ,..., X _{k + 2N−1} } correlation coefficient c _a of interest to have a pitch waveform with the pitch waveforms of the previous section are determined using the equations shown in equation 3, attention to that the correlation between the pitch waveforms and the pitch waveforms subsequent sections the number c _b is determined using the equation shown in formula 4. Control unit 110, c _a, pitch waveform corresponding to the larger of the values of c _b is determined that there is high correlation than the other pitch waveform is selected pitch waveforms having a higher correlation.

  The equations shown in Equations 3 and 4 take the cross-correlation of the two pitch waveforms, but these pitch waveforms are originally extracted from the same speech waveform data. Therefore, in the end, the equations shown in Equations 3 and 4 take the autocorrelation of the speech waveform data.

In the case of waveform reduction, the waveform reduction / expansion unit 114 generates a pitch waveform from a pitch waveform of interest and a pitch waveform having a higher correlation coefficient with the pitch waveform. The procedure for generating the pitch waveform will be described later. Then, an operation of replacing the pitch waveform of interest with the generated pitch waveform and the pitch waveform having a higher correlation coefficient with the pitch waveform is performed.

In the case of waveform expansion, the waveform reduction / expansion unit 114 generates a pitch waveform from a pitch waveform of interest and a pitch waveform having a higher correlation coefficient with the pitch waveform. The procedure for generating the pitch waveform will be described later. Then, an operation of inserting the generated pitch waveform between the pitch waveform of interest and a pitch waveform having a higher correlation coefficient with the pitch waveform is performed.

  Returning to FIG. 1, for example, the input control unit 120 connects the input device 12 such as a keyboard or a pointing device, receives an instruction to the control unit 110 input from the input device 12, and transmits the instruction to the control unit 110. .

  For example, the output control unit 130 connects the output device 13 such as a display or a speaker, and outputs the processing result of the control unit 110 to the output device 13 as necessary.

  The program storage unit 140 is configured by a ROM (Read Only Memory) or the like, and stores an operation program executed by the control unit 110.

  The storage unit 150 is composed of a storage device such as a hard disk device or RAM (Random Access Memory), for example, and the audio waveform data sent from the data recording unit 170 and the audio waveform after the waveform expansion process or the waveform reduction process. Temporarily store data. The storage unit 150 sends the temporarily stored audio waveform data to the data recording unit 170 or the control unit 110.

  The data recording unit 170 is, for example, a CD-RW drive or the like, and reads audio waveform data stored in the recording medium 17 in accordance with an instruction from the control unit 110. Further, the expanded or reduced audio waveform data is written to the recording medium 17.

  The waveform reduction / expansion processing will be described below with reference to the drawings. FIG. 3 is a flowchart of the waveform reduction / expansion processing. In the following description, the reduction / expansion magnification a is between 1/2 and 2.

First, the control unit 110 calculates a target reduction / expansion length Fa based on the designated reduction / expansion magnification a (step S101). The formula for calculating Fa is shown in Equation 5.
(Equation 5)
Fa = F × a−F
Here, F is the length (number of samples) of the audio frame. When the input waveform is reduced, Fa becomes negative, and when it expands, Fa becomes positive.

  Next, the dividing unit 111 in the control unit 110 divides the input waveform into audio frames having a length F (step S102). Then, the first audio frame is set as a focused audio frame.

  The controller 110 determines whether or not the input waveform to be processed remains (step S103), and determines that the input waveform to be processed does not remain and the entire input waveform has been reduced or expanded (step S103: NO). ), The waveform reduction / expansion processing is terminated. When it is determined that the input waveform to be processed remains (step S103: YES), the control unit 110 repeats the following steps S104 to S112 until there is no input waveform to be processed.

In step S104, the pitch extraction unit 112 cuts out a pitch waveform from the audio frame by the method described above. Then, this is transferred to the pitch selection unit 113. The pitch selection unit 113 calculates the similarity of each pitch waveform ( correlation coefficient with an adjacent pitch waveform ) (step S105, the calculation formula is Equation 2). The pitch selection unit 113 determines the processing order of the pitch waveforms so that the high-similarity pitch waveform calculated in step S105 is first reduced or expanded (step S106).

  The waveform reduction / expansion unit 114 performs reduction or expansion of the pitch waveform in the processing order determined by the pitch selection unit 113. First, the waveform reduction / expansion unit 114 selects a pitch waveform in the determined processing order (step S107). Next, it is determined whether or not the accumulated processing within the audio frame being processed exceeds the absolute value of Fa (step S108).

  When it is determined that the accumulated processing of pitch waveforms in the speech frame (total length of speech waveforms generated by expansion / reduction) exceeds the absolute value of Fa (step S108: YES), the waveform reduction / expansion unit In step S109, the difference 114 between the accumulated process and the absolute value of Fa is carried over to the next process. That is, in the processing of the second audio frame, in step S108, it is determined whether or not the accumulated processing of the pitch waveform in the audio frame has exceeded 2 | Fa |-(the accumulated processing of the previous pitch waveform). Thereafter, the accumulated processing of the pitch waveform so far affects the next processing range, and the value compared with the accumulated processing of the current pitch waveform varies in step S108. When the process of step S109 ends, the process moves to step S103, and the next audio frame is processed.

  When it is determined that the accumulated processing of the pitch waveform within the audio frame does not exceed the absolute value of Fa (step S108: NO), the waveform reduction / expansion unit 114 determines whether or not the designated magnification a is 1 or more. (Step S110), if it is 1 or more (step S110: YES), the expansion process (step S111) is executed for the pitch waveform. If it is less than 1 (step S110: NO), the pitch waveform is processed. A reduction process (step S111) is executed. And the control part 110 returns a process to step S107.

  The decompression process in step S111 will be described with reference to the flowchart in FIG.

Hereinafter, each sample value sequence of the pitch waveform of interest (designated to expand the waveform) is _represented as {x _k , x _{k + 1} ,..., X _{k + N−1} }, and the pitch before this pitch waveform Each sample value sequence of the waveform is _represented by {x _k−N , x _{k−N + 1} ,..., X _k−1 }, and each sample value sequence of the pitch waveform after the pitch waveform of interest is _represented by {x _{k + N} , X _{k + N + 1} ,..., X _{k + 2N−1} }.

Next, the waveform reduction / expansion unit 114 calculates _a correlation coefficient ca between the pitch waveform of interest and the pitch waveform of the previous section using the equation shown in Formula 3, and calculates the pitch waveform of interest. calculated using the equation shown in equation (4) the correlation coefficient c _b between the pitch waveforms of the subsequent interval (Figure 4: step S301).

Then, the waveform reduction / expansion unit 114 determines the magnitude of c _a and c _b calculated in step S301, the correlation between the pitch waveforms of interest to determine the higher pitch waveform (step S302).

When the correlation of the past-side pitch waveform is higher than the correlation of the future-side pitch waveform (step S302: past-side (previous)), the waveform reduction / expansion unit 114 generates a pitch waveform according to the following equation (6). (Step S303).
(Equation 6)
s _i = (i / N−1) × x _{k−N + i} + ((N−1−i) / N−1) × x _{k + i}
(I is 0 to N-1)

  The expression shown in Expression 6 indicates that the sample values of the past-side pitch waveform and the pitch waveform of the section of interest are weighted and added. The weight coefficient (i / N-1) of the pitch waveform on the past side starts from 0 and ends with 1. The weighting coefficient ((N-1-i) / N-1) of the pitch waveform in the attention section starts with 1 and ends with 0.

  Next, the waveform reduction / expansion unit 114 connects (inserts) the generated pitch waveform between the previous pitch waveform and the pitch waveform of interest (step S304), ends the expansion process, and proceeds to step S107.

The respective sample values of the speech waveform obtained by the processing of the above steps are {..., X _k−1 , s ₀ , s ₁ ,..., S _N−1 , x _k , x _{k + 1} , , X _{k + N−1} ,.

On the other hand, when the correlation of the pitch waveform on the future side is higher than the correlation of the pitch waveform on the past side (FIG. 4: step S302: future side (after)), the waveform reduction / expansion unit 114 follows the equation shown in the following equation (7). A pitch waveform is generated (step S305).
(Equation 7)
s _i = (i / N−1) × x _{k + i} + ((N−1−i) / N−1) × x _{k + N + i}
(I is 0 to N-1)

  Next, the waveform reduction / expansion unit 114 connects (inserts) the generated pitch waveform between the focused pitch waveform and the subsequent pitch waveform (step S306), and ends the expansion process.

Each sample value of the speech waveform obtained by the processing of the above steps is {..., X _k , x _{k + 1} ,..., X _{k + N−1} , s ₀ , s _{1 ,. 1} , x _{k + N} ,.

  The reduction process in step S112 will be described with reference to the flowchart of FIG.

  Steps S401 and S402 are the same as steps S301 and S302 in FIG.

In step S403, as in step S303 of FIG. 4, the waveform reduction / expansion unit 114 performs weighted addition using the equation shown in Equation 8 based on the pitch waveform of interest and the previous pitch waveform, and creates a new pitch. Generate a waveform. As is clear from the comparison between Equation 6 and Equation 8, the weighting factor is the reverse of the expansion. Therefore, the weight coefficient (i / N-1) of the pitch waveform on the past side starts from 1 and ends at 0. The weighting coefficient ((N-1-i) / N-1) of the pitch waveform in the attention section starts with 0 and ends with 1.
(Equation 8)
s _i = ((N−1−i) / N−1) × x _{k−N + i} + (i / N−1) × x _{k + i}
(I is 0 to N-1)

  Then, the waveform reduction / expansion unit 114 replaces the pitch waveform of interest and the previous pitch waveform with the generated pitch waveform (step S404). That is, two consecutive pitch waveforms are replaced with generated pitch waveforms.

In step S405, as in step S305 of FIG. 4, the waveform reduction / expansion unit 114 performs weighted addition using the equation shown in Equation 9 based on the pitch waveform of interest and the subsequent pitch waveform, and creates a new pitch. Generate a waveform. Then, the pitch waveform of interest and the subsequent pitch waveform are replaced with the generated pitch waveform (step S406).
(Equation 9)
s _i = ((N−1−i) / N−1) × x _{k + i} + (i / N−1) × x _{k + N + i}
(I is 0 to N-1)

  According to such a configuration, the pitch waveform is generated based on the pitch waveform having the higher correlation between the past side and the future side and the pitch waveform of interest. In the case of expansion, it is inserted immediately before or after the pitch waveform of interest. In the case of reduction, the generated pitch waveform is inserted instead of the voice waveform. For this reason, it is possible to reduce the generation of noise when reproducing a transitional speech waveform. Of the two pitch waveforms, the past-side pitch waveform is multiplied by a weighting factor starting from 0 and ending with 1, and the future-side pitch waveform is weighted starting with 1 and ending with 0. Multiply. For this reason, the generated pitch waveform is connected to the preceding and succeeding pitch waveforms while maintaining the continuity of the waveform.

  Note that when the input speech waveform is reduced and output, since it is replaced by weighted addition with another pitch waveform, there is a pitch waveform that is erased before the processing order comes around. In this case, such a pitch waveform is not a processing target in step S112.

Further, the weighting coefficients of the above-described equations 6 to 9 are examples, and N number sequences a _i (i / (N−1) in the first embodiment) and 1 starting from 0 and ending with 1 and 1 Any number may be used as long as it is an N number sequence b _i that starts and ends with 0 ((N-1-i) / (N-1) in the first embodiment). However, for each i (from 0 to N−1), it is necessary to satisfy the relationship expressed by the following equation (10).
(Equation 10)
a _i + b _i = 1

  The pitch extraction unit 112 sets the remaining section as the head of the next process when the remaining section is equal to or less than the specified multiple, but starts the next process ( Pitch waveform detection) may be performed. When processing is performed in this way, processing is executed in units shorter than the above processing method. For this reason, an output closer to the designated magnification can be obtained.

  Further, as a condition for stopping the cutout of the pitch waveform, the cutout pitch waveform having the largest number of samples (maximum value) was used as a reference. This is the arithmetic average of the number of samples (hereinafter simply referred to as the average). It can be. In this case, it is possible to avoid the case where the length of the pitch waveform is erroneously determined and the process is unnecessarily transferred to the next process.

  Although the case where the magnification is between 1/2 and 2 has been described so far, the magnification may be outside this range.

For example, when the magnification is between n times and (n + 1) times (where n is an integer greater than or equal to 2 and does not include n times, but includes (n + 1) times), the pitch waveform selected according to the processing order Undergoes (n + 1) times expansion processing. When receiving the (n + 1) times expansion process, in step S302, the waveform reduction / expansion unit 114 selects a pitch waveform for generating an expansion waveform as follows. Here, the head position of the pitch waveform of interest is set to 0, and the length of the pitch waveform is set to N.
1) When past side correlation is greater than future side correlation-Speech waveform from nN to 0 and speech waveform from 0 to nN 2) When future side correlation is greater than past side correlation (1-n) N Speech waveform from N to N and speech waveform from N to (1 + n) N

The weighting coefficients of Equations 6 and 7 are i / (nN-1) instead of i / (N-1), and (nN-) instead of (N-1-i) / (N-1). 1-i) / (nN-1). The correlation coefficient c _a is determined from the speech waveform from −nN to 0 and the speech waveform from 0 to nN, and the correlation coefficient c _b is determined from the speech waveform from (1-n) N to N and from N. It is desirable to obtain from speech waveforms up to (1 + n) N.

Also, for example, when the magnification is between 1 / (m + 1) times and 1 / m times (where m is an integer of 2 or more and does not include 1 / m times, but includes 1 / (m + 1) times) ), The selected pitch waveform is subjected to a reduction process of 1 / (m + 1) times. When the 1 / (m + 1) times reduction process is performed, in step S402, the waveform reduction / expansion unit 114 selects a pitch waveform for generating a reduced waveform as follows. Here, the head position of the pitch waveform of interest is set to 0, and the length of the pitch waveform is set to N.
1) When the correlation on the past side is larger than the correlation on the future side The speech waveform from -mN to (-m + 1) N and the speech waveform from 0 to N 2) When the correlation on the future side is greater than the correlation on the past side From 0 Speech waveform up to N and speech waveform from mN to (m + 1) N

In step S404, the generated pitch waveform replaces the speech waveform from the previous pitch waveform to the target pitch waveform. In step S406, the generated pitch waveform replaces the target pitch waveform to the subsequent pitch waveform.

At this time, the correlation coefficient c _a is a speech waveform from −mN to 0 and a speech waveform from (−m + 1) N to N, and the correlation coefficient c _b is a speech waveform from 0 to mN. It is desirable to obtain from speech waveforms from N to (m + 1) N.

(Embodiment 2)
In the first embodiment, the pitch waveform similarity is obtained, and the pitch waveform is expanded / reduced in descending order of similarity. However, since the pitch waveform is reduced / expanded based only on the measure of similarity, the pitch waveform is reduced / expanded regardless of the phoneme language importance. For this reason, for example, when a human conversation is played back at high speed, the pitch waveform on which stress (stress, accent) is placed is also reduced, which causes a problem that it becomes difficult for humans to hear the conversation. Therefore, in the second embodiment, it is determined whether or not there is a portion (hereinafter, referred to as a stress portion) in which a stress is placed in the audio frame, and the playback time is not shortened as much as possible. A speech processing apparatus according to the second embodiment will be described.

  Since the configuration of the speech processing apparatus 100 according to the present embodiment is the same as that of the first embodiment, only differences from the first embodiment will be described.

  When the reduction of the audio signal is instructed, the pitch selection unit 113 of the present embodiment further determines whether or not there is a stress portion when ordering the sound frames divided for each pitch, and determines the stress portion. The processing order of the pitch including it is postponed.

  The pitch selection unit 113 performs the determination according to the following procedure in order to determine whether or not there is a stressed portion in the audio frame.

  First, the pitch waveforms are once ordered by the method described in the first embodiment (arranged in descending order of similarity). Since a pitch waveform that resembles the front and rear appears in the portion where the stress is placed, the pitch waveform including the stress portion is placed higher in this ordering. Further, since the amplitude is larger than the others, the stress portion has a larger wave energy than the other portion (pitch waveform).

  Therefore, the pitch selection unit 113 determines whether or not there is a pitch waveform including a stress portion from the state of change of the energy value of each ordered pitch waveform (hereinafter referred to as “stress determination processing”). ). If it is determined that there are strong portions, the rank of those portions is lowered to a predetermined order.

  The stress determination process will be described with reference to the flowchart shown in FIG. The stress determination process is executed after step S106 in FIG. 3 and before step S107.

  First, the pitch selection unit 113 substitutes 1 for a variable counter i as an initialization process (step S601).

Next, the value of i is compared with the value of N / t, and it is determined whether or not the value of i is smaller than the value of N / t (step S602). If the value of i is smaller than the value of N / t (step S602: YES), the pitch selection unit 113 continues to execute the stress determination process, and moves the process to step S603. N is the number of pitch waveforms to be processed in the stress determination process. Further, t is a value obtained experimentally, and is 4 in this embodiment.

  If the value of i is larger than the value of N / t (step S602: NO), the pitch selection unit 113 determines that there is no stress in the audio frame (step S607), and ends the stress determination process. In this case, there is no change in the order more than that performed in step S106.

In step S603, k ₀ and k ₁ are calculated by the equations shown in equations 11 and 12. Note that sw _j is an energy value of a wave included in the pitch waveform corresponding to the processing order j given in step S106 described above. sw _j is calculated by the equation shown in Equation 13. S _x , Pl _n , and _pn are the same as the definitions used in Equation 2. Further, n in Equation 13 means that the pitch waveform corresponding to the processing order j is the nth pitch waveform from the beginning of the audio frame.

If the stress portion is included in the voice frame, the amplitude of the stress portion is larger than the amplitude of the other portions. Therefore, if the frequency hardly changes in the voice frame, the energy of the pitch waveform including the stress portion is particularly higher than the energy of the other pitch waveforms not including the stress portion. Therefore, the value of k ₀ is compared with the constant multiple b of k ₁ (step S604). If the value of k ₀ is equal to or less than the constant multiple b of k ₁ (step S604: NO), 1 is added to i. Then (step S605), the process returns to step S602. Note that the value of b is an experimentally obtained optimum value, for example, 1.5.

When it is determined that the value of k ₀ is larger than the constant multiple b of k ₁ (step S604: YES), the pitch selection unit 113 determines that there is a stress in the audio frame (step S606). Then, the processing order of descending order in i pieces took pitch waveform similarity contained in k _0, changing to receive reduced later than the other pitch waveforms. The value of i here is the value of i when the value of k ₀ becomes larger than the constant multiple b of k ₁ in step S604. Then, the stress determination process ends.

  However, if the processing order of the pitch waveform having the stress portion is lowered last, the pitch waveform that is originally not suitable for the reduction processing is also reduced, and the sound quality may be further deteriorated. Therefore, the pitch selection unit 113 arithmetically averages the processing order of the pitch waveform, for example, the maximum similarity value and the maximum similarity value, and the processing order given to the pitch waveform having the degree of similarity of the obtained values. Deterioration of sound quality can be suppressed by not lowering. If there is no pitch waveform having a similarity with a value equal to the arithmetic average value, the processing order is higher than the processing order assigned to the pitch waveform having the largest value among the pitch waveforms having a similarity lower than the arithmetic mean value. Shall not be lowered.

  With such a configuration, it is difficult to reduce the voice waveform of a strong portion, and for example, it is possible to improve the ease of hearing during high-speed playback. In addition, since the processing order of the strong portion is not lowered below a certain order, it is possible to suppress deterioration in sound quality during reproduction.

  As described above, in the speech processing apparatus 100 according to the first and second embodiments, the input speech waveform data can be reduced or expanded at an arbitrary magnification and output.

  In addition, this invention is not limited to the said embodiment, A various deformation | transformation and application are possible.

For example, in the above embodiment, the angle of the average error v _n and a vector shown in Equation 15, but using the least square error expressed by Equation 2 as a method for selecting the location where pitch waveforms similar to continuous, as shown in Equation 14 The coefficient h _n may be used. Pitch selection unit 113, when using the average error v _n is ascending order of average error, when using the angular coefficient h _n selects a waveform to shrink / stretch in descending order of angular coefficient. In the case of the second embodiment, after the order of pitch waveforms is rearranged by the equation shown in Equation 13 or 14, the above-described stress determination process is executed, and the order of pitch waveforms is further rearranged.

Ｓ_ｘ、Ｐｌ_ｎ、ｐ_ｎの定義は、数１４、数１５ともに数２と同じである。

Definition of S _{x, Pl n, p} n is the number 14, is the same as the number 15 together several 2.

  In the above equation 2, the mean square error between the pitch waveform of interest and the pitch waveform positioned before the pitch waveform on the input waveform is calculated, but the pitch waveform positioned after the pitch waveform on the input waveform is calculated. The mean square error may be calculated. Further, an average square error with the pitch waveform positioned before and after the pitch waveform on the input waveform may be calculated, and one of them (larger or smaller) may be adopted as a representative value of similarity.

  Further, the audio processing apparatus according to each of the above embodiments may be configured to process only one of expansion and reduction.

  In addition, the audio processing device according to each of the above embodiments may further include a communication control unit that communicates with another device via a network such as the Internet, and the audio waveform data is transmitted via the communication control unit. You may make it transmit / receive with another apparatus.

  In addition, the voice processing apparatus 100 may accept an analog voice input. In this case, the audio processing apparatus 100 further includes an audio sampling unit that samples analog audio data by a predetermined method such as PCM (Pulse Code Modulation). Further, the sound processing apparatus 100 may output analog sound.

  Note that the audio processing apparatus 100 in each of the above embodiments can be realized using a normal computer system, not a dedicated system. For example, a program for executing the above operation is stored in a computer-readable recording medium (FD, CD-ROM, DVD, etc.) and distributed, and the program is installed in the computer to execute the above processing. The audio processing / playback apparatus 100 may be configured. Alternatively, it may be stored in a disk device of a server device on a network such as the Internet, and downloaded to a computer, for example.

  In addition, when the OS realizes the above functions by sharing the OS or jointly of the OS and the application, only the part other than the OS may be stored and distributed in the medium, or may be downloaded to the computer. Good.

It is a block diagram of the audio processing apparatus concerning embodiment of this invention. It is a logic block diagram of the control part of FIG. It is a flowchart for demonstrating the waveform reduction / expansion process concerning embodiment of this invention. It is a flowchart for demonstrating the expansion | extension process concerning embodiment of this invention. It is a flowchart for demonstrating the reduction process concerning embodiment of this invention. It is a flowchart for demonstrating the stress discrimination | determination process concerning Embodiment 2 of this invention. It is a principle explanatory drawing of the TDHS system which is a conventional audio compression / reduction system.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 ... Voice processing apparatus, 110 ... Control part, 111 ... Dividing part, 112 ... Pitch extraction part, 113 ... Pitch selection part, 114 ... Waveform reduction / expansion part, 120 ... Input control part, 12 ... Input device, 130 ... Output Audio processing unit, 13 ... output device, 140 ... program storage unit, 150 ... storage unit, 170 ... external storage IO device, 17 ... storage medium

Claims (9)

A speech waveform expansion device for expanding and outputting an input waveform on a time axis,
Input waveform receiving means for receiving data representing the input waveform;
An expansion ratio acceptance means for accepting an input of a magnification for expanding the input waveform;
Cutting means for cutting out a pitch waveform from the input waveform received by the input waveform receiving means;
For each pitch waveform cut out by the cut-out means, similarity calculation means for calculating the similarity between the pitch waveform and one pitch waveform adjacent to the pitch waveform before and after on the time axis; ,
Ordering means for assigning a processing order to each pitch waveform using the similarity calculated by the similarity calculation means as a scale;
Expanded waveform generation that generates a waveform for insertion by selecting a pitch waveform in the processing order given by the ordering means and weighting and adding the selected pitch waveform and one pitch waveform adjacent to the pitch waveform Means,
An expanded waveform connecting means for inserting the waveform generated by the expanded waveform generating means between two pitch waveforms that are subjected to weighted addition on the input waveform;
Equipped with,
Repeating the processing by the expanded waveform generating means and the expanded waveform connecting means while updating the pitch waveform to be selected according to the processing order until the waveform length of the specified magnification is reached.
Speech waveform decompression apparatus according to claim. The cutout means is means for cutting out a partial waveform of a certain length from the input waveform, and cutting out a pitch waveform in units of the cutout partial waveform,
The ordering means performs processing of ordering pitch waveforms in units of the fixed-length partial waveform.
The speech waveform expansion device according to claim 1. An input waveform receiving step for receiving data representing the input waveform;
An expansion magnification reception step for receiving an input of a magnification for expanding the input waveform;
A step of cutting out a pitch waveform from the input waveform received in the input waveform receiving step;
For each pitch waveform cut out in the cut-out step, a similarity calculation step for calculating a similarity between the pitch waveform and one pitch waveform adjacent to the pitch waveform before and after on the time axis; ,
An ordering step of assigning a processing order to each pitch waveform using the similarity calculated in the similarity calculation step as a scale;
Choose Back Symbol ordered pitch waveforms applied treatment sequence in step, by weighted addition of the one pitch waveform adjacent to the selected pitch waveform and the pitch waveform, decompression waveform to generate a waveform for insertion Generation step;
An expanded waveform connecting step for inserting the waveform generated in the expanded waveform generating step between two pitch waveforms that are subjected to weighted addition on the input waveform;
With
Until waveform length of the specified magnification, while updating the pitch waveforms to be selected according to the processing order, characterized said expansion waveform generation step, and a score repeat processing by said expansion waveform connecting step Waveform stretching method. The computer used in the speech waveform decompression apparatus for decompressing and outputting input waveform on the time axis,
Input waveform receiving means for receiving data representing the input waveform ;
And stretching magnification accepting means for accepting an input of a magnification stretching the input waveform,
And the cut-out means that you cut out the pitch waveform from the input waveform received by the input waveform receiving means,
For each pitch waveform cut out by the cut- out means, similarity calculation means for calculating the similarity between the pitch waveform and one pitch waveform adjacent to the pitch waveform before and after on the time axis; ,
Ordering means for assigning a processing order to each pitch waveform using the similarity calculated by the similarity calculation means as a scale;
Expanded waveform generation that generates a waveform for insertion by selecting a pitch waveform in the processing order given by the ordering means and weighting and adding the selected pitch waveform and one pitch waveform adjacent to the pitch waveform Means ,
An expanded waveform connecting means for inserting the waveform generated by the expanded waveform generating means between two pitch waveforms that are subjected to weighted addition on the input waveform;
To function,
Until waveform length of the specified magnification, while updating the pitch waveforms to be selected according to the processing order, and said decompressed waveform generating means, that it has the decompressed waveform connecting means and repeatedly to by the processing returns Suyo A featured program. A speech waveform reduction device that reduces and outputs an input waveform on a time axis,
Input waveform receiving means for receiving data representing the input waveform;
A reduction magnification receiving means for receiving an input of a magnification for reducing the input waveform;
Cutting means for cutting out a pitch waveform from the input waveform received by the input waveform receiving means;
For each pitch waveform cut out by the cut-out means, similarity calculation means for calculating the similarity between the pitch waveform and one pitch waveform adjacent to the pitch waveform before and after on the time axis; ,
Ordering means for assigning a processing order to each pitch waveform using the similarity calculated by the similarity calculation means as a scale;
Reduced waveform generation that generates a replacement waveform by selecting a pitch waveform in the processing order given by the ordering means and weighting and adding the selected pitch waveform and one pitch waveform adjacent to the pitch waveform Means,
The waveform generated by the reduced waveform generating means, and two reduced waveform connecting means Ru replaced with pitch waveform subject to weighting addition in the reduced waveform generating means on the input waveform,
Equipped with,
Repeating the processing by the reduced waveform generating means and the reduced waveform connecting means while updating the pitch waveform to be selected according to the processing order until the waveform length of the specified magnification is reached.
Speech waveform reduction apparatus according to claim. The cutout means is means for cutting out a partial waveform of a certain length from the input waveform, and cutting out a pitch waveform in units of the cutout partial waveform,
The ordering means performs processing of ordering pitch waveforms in units of the fixed-length partial waveform.
The speech waveform reduction device according to claim 5 . An input waveform reception step for receiving data representing the input waveform;
A reduction magnification acceptance step for accepting an input of a magnification for reducing the input waveform;
A step of cutting out a pitch waveform from the input waveform received in the input waveform receiving step;
For each pitch waveform cut out in the cut-out step, a similarity calculation step for calculating a similarity between the pitch waveform and one pitch waveform adjacent to the pitch waveform before and after on the time axis; ,
An ordering step of assigning a processing order to each pitch waveform using the similarity calculated in the similarity calculation step as a scale;
Choose Back Symbol ordered pitch waveform processing order granted in step, by weighted addition of the one pitch waveform adjacent to the selected pitch waveform and the pitch waveform, reduced to generate a waveform for replacement waveform Generation step;
The waveform generated by the reduced waveform generation step, and two reduced waveform connecting step of Ru replaced with pitch waveform subject to weighting addition in the reduced waveform generation step on the input waveform,
With
Until waveform length of the specified magnification, while updating the pitch waveforms to be selected according to the processing order, characterized with the reduced waveform generation step, and a score repeat the processing of and the reduced waveform connecting step Waveform reduction method. The computer used to input waveform to the sound wave reduction device for outputting reduced on the time axis,
Input waveform receiving means for receiving data representing the input waveform ;
A reduction ratio accepting means for accepting an input of a magnification reducing the input waveform,
And the cut-out means that you cut out the pitch waveform from the input waveform received by the input waveform receiving means,
For each pitch waveform cut out by the cut- out means, similarity calculation means for calculating the similarity between the pitch waveform and one pitch waveform adjacent to the pitch waveform before and after on the time axis; ,
Ordering means for assigning a processing order to each pitch waveform using the similarity calculated by the similarity calculation means as a scale;
Reduced waveform generation that generates a replacement waveform by selecting a pitch waveform in the processing order given by the ordering means and weighting and adding the selected pitch waveform and one pitch waveform adjacent to the pitch waveform Means ,
The waveform generated by the reduced waveform generating means, a reduced waveform connecting means for replacing two pitch waveforms subject to weighting addition in the reduced waveform generating means on the input waveform,
To function,
Until waveform length of the specified magnification, while updating the pitch waveforms to be selected according to the processing order, and the reduced waveform generating means, that it has to the reduced waveform connecting means and repeatedly to by the processing returns Suyo A featured program. A speech processing apparatus that outputs an input waveform by extending or reducing on a time axis,
Input waveform receiving means for receiving data representing the input waveform;
Magnification accepting means for accepting an input of a magnification for expanding or reducing the input waveform;
Cutting means for cutting out a pitch waveform from the input waveform received by the input waveform receiving means;
For each pitch waveform cut out by the cut-out means, similarity calculation means for calculating the similarity between the pitch waveform and one pitch waveform adjacent to the pitch waveform before and after on the time axis; ,
Ordering means for assigning a processing order to each pitch waveform using the similarity calculated by the similarity calculation means as a scale;
Expanded waveform generation that generates a waveform for insertion by selecting a pitch waveform in the processing order given by the ordering means and weighting and adding the selected pitch waveform and one pitch waveform adjacent to the pitch waveform Means,
An expanded waveform connecting means for inserting the waveform generated by the expanded waveform generating means between two pitch waveforms that are subjected to weighted addition on the input waveform;
Reduced waveform generation that generates a replacement waveform by selecting a pitch waveform in the processing order given by the ordering means and weighting and adding the selected pitch waveform and one pitch waveform adjacent to the pitch waveform Means,
The waveform generated by the reduced waveform generating means, and two reduced waveform connecting means Ru replaced with pitch waveform subject to weighting addition in the reduced waveform generating means on the input waveform,
Magnification discrimination means for discriminating whether to expand or reduce the input waveform;
When it is determined that the input waveform is to be expanded by the magnification determining means, the expanded waveform generating means is updated while updating the pitch waveform selected according to the processing order until the waveform length of the specified magnification is reached. And first repeating means for repeating the processing by the expanded waveform connecting means,
When it is determined that the input waveform is to be reduced by the magnification determining means, the reduced waveform generating means is updated while updating the pitch waveform selected according to the processing order until the waveform length of the specified magnification is reached. And second repeating means for repeating the processing by the reduced waveform connecting means,
Speech processing apparatus characterized by comprising a.

Images