JP5228300B2 - Audio expansion device, audio expansion method, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To reduce degradation of voice quality by voice extension and contraction processing. <P>SOLUTION: A voice processing device 1101 extends and contracts a voice signal with magnification instructed via an input device 1111. Extension and contraction are applied for each predetermined unit waveform. A control section 1121 puts a priority order for performing insertion of a new unit waveform or replacing to the new unit waveform, for each predetermined unit waveform, and according to the order, the insertion or the replacing is performed, and thereby, the voice signal is gradually extended and contracted. The new unit waveform is created based on an original unit waveform, however the same unit waveform is not made continuous so that the voice after extension and contraction processing may not become unnatural. <P>COPYRIGHT: (C)2008,JPO&amp;INPIT

Description

The present invention relates to an audio expansion device , an audio expansion method , and a program that process an audio signal in the time domain to expand and contract the audio signal.

  As one of the processes for transforming the audio signal, there is a process for expanding or contracting the length (number of samples) of the audio signal to be processed without changing the amplitude or frequency characteristic of the audio signal to be processed.

  For example, in a scene where a portion difficult to hear in an English conversation teaching material is reproduced slowly, an audio signal expansion process is necessary.

  On the other hand, in a scene where it is desired to grasp the contents of a recorded meeting or lecture in a short time, it is necessary to reduce the audio signal.

  In order to expand an audio signal, a method of generating a new partial waveform based on two consecutive partial waveforms and interrupting the new partial waveform between the two consecutive partial waveforms is disclosed (patent). Reference 1). Hereinafter, a partial waveform obtained by dividing an audio signal according to some rules may be referred to as a divided waveform. For example, a method using a pitch waveform as a unit is disclosed as a method of such separation (Patent Document 2).

  In order to extend the speech waveform by inserting a new segmented waveform, the simplest method is to use a copy of one of the two consecutive segmented waveforms as it is as a new segmented waveform.

  However, if this is done, the decompressed speech will be unnatural speech because there is a time zone in which exactly the same segmental waveform continues.

  This disadvantage can be eliminated if the new segment waveform is an intermediate waveform of segment waveforms before and after the new segment waveform is inserted, for example, a weighted average waveform.

  However, it is necessary to be careful when the degree of expansion exceeds twice. When the degree of expansion exceeds twice, a portion that requires interrupting a plurality of new speech waveforms appears at the boundary between two consecutive segmented waveforms in the original speech signal. In this case, when the plurality of new speech waveforms are the same, the decompressed speech eventually becomes unnatural speech because there is a time zone in which exactly the same segmental waveform continues.

  On the other hand, in order to reduce the audio signal, the simplest method is to thin out the segmented waveforms as appropriate.

However, if the segmented waveforms are simply thinned out, the segmented waveforms that are not continuous in the original audio signal are newly adjacent. Then, since the waveform change of the audio signal becomes abrupt in the adjacent portion, the reduced audio becomes unnatural audio.
JP 2005-275010 A JP-A-9-258762

  As described above, at the time of voice decompression, there is a concern that the voice quality is deteriorated due to a plurality of continuous identical waveforms. On the other hand, at the time of audio reduction, there is a concern that the audio quality is deteriorated due to a sudden change in the audio signal.

The present invention has been made in view of the above problems, and an object of the present invention is to provide an audio expansion device, an audio expansion method, and a program that reduce deterioration of an audio waveform even when the expansion degree exceeds twice .

The audio decompression apparatus according to the present invention is
A storage unit for storing a speech waveform that is a time series of segmented waveforms;
With respect to two consecutive segment waveforms in the speech waveform stored in the storage unit, a weighting factor that increases linearly in time with respect to the front segment waveform and time with respect to the rear segment waveform. Each using a weighting factor that decreases linearly in each case, and a new segmented waveform generating means for generating one new segmented waveform by weighted addition ,
A plurality of insertions each having a different length between the length of the front segment waveform and the length of the rear segment waveform by expanding / contracting the new segment waveform generated by the new segment waveform generation means on the time axis Insertion segment waveform generation means for generating segment waveform,
Insertion means for inserting a plurality of insertion segment waveforms generated by the insertion segment waveform generation unit between the front segment waveform and the rear segment waveform;
A voice waveform output means for outputting a voice waveform expanded by inserting an insertion section waveform by the insertion means;
It is characterized by comprising.

A speech decompression method according to the present invention is a speech decompression method for decompressing a speech waveform that is a time series of segmented waveforms,
For two consecutive segment waveforms in the speech waveform, a weighting factor that increases linearly in time for the preceding segment waveform and decreases linearly in time for the rear segment waveform. A new segment waveform generation step for generating one new segment waveform by weighting and adding the weighting factors to be used, respectively,
A plurality of insertions having different lengths between the length of the front segment waveform and the length of the rear segment waveform by expanding and contracting the new segment waveform generated in the new segment waveform generation step on the time axis. An insertion segment waveform generation step for generating a segment waveform ; and
An insertion step of inserting a plurality of insertion segment waveforms generated in the insertion segment waveform generation step between the front segment waveform and the rear segment waveform;
It is characterized by comprising.

The program according to the present invention is:
A computer of a voice expansion device that expands a voice waveform that is a time series of segmented waveforms stored in a storage unit,
With respect to two consecutive segment waveforms in the speech waveform stored in the storage unit, a weighting factor that increases linearly in time with respect to the front segment waveform and time with respect to the rear segment waveform. Each newly using a weighting factor that decreases linearly, and a new segmented waveform generating means for generating one new segmented waveform by weighted addition,
A plurality of insertions each having a different length between the length of the front segment waveform and the length of the rear segment waveform by expanding / contracting the new segment waveform generated by the new segment waveform generation means on the time axis Insertion section waveform generation means for generating section waveforms,
Insertion means for inserting a plurality of insertion segment waveforms generated by the insertion segment waveform generation unit between the front segment waveform and the rear segment waveform;
A voice waveform output means for outputting a voice waveform expanded by inserting an insertion section waveform by the insertion means;
It is made to function as.

According to the present invention, it is possible to reduce degradation of a speech waveform after decompression even when the speech waveform is stretched by a factor of two or more .

(Embodiment 1)
Embodiment 1 according to the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of a speech processing apparatus, particularly a speech decompression apparatus, according to Embodiment 1 of the present invention.

  As shown in FIG. 1, the voice processing device 1101 is configured by an information processing device such as a computer, for example. An input device 1111, an output device 1113, and a recording medium 1115 are connected to the sound processing device 1101. In response to an instruction from the input device 1111, the audio processing device 1101 expands the audio waveform data input from the recording medium 1115 to a specified multiple length, and outputs it to the recording medium 1115.

  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 1115 is, for example, a CD-RW (Compact Disk ReWritable) disk or the like. The recording medium 1115 stores the original audio waveform data or the audio waveform data obtained by expanding the audio waveform data to a specified multiple length. Or store.

  The voice processing device 1101 includes a control unit 1121, an input control unit 1131, an output control unit 1133, a program storage unit 1135, a storage unit 1137, and a data recording unit 1139.

  The control unit 1121 includes, for example, a CPU (Central Processing Unit), a register, a RAM (Random Access Memory), and the like, and is based on an operation program stored in advance in the program storage unit 1135. 1101 is controlled, the audio waveform data stored in the recording medium 1115 is read out via the data recording unit 1139, the expanded audio waveform data is written in the recording medium 1115, a waveform expansion process to be described later, and the like. Or execute.

  The input control unit 1131 connects, for example, an input device 1111 such as a keyboard or a pointing device, receives an instruction to the control unit 1121 input from the input device 1111, and transmits the instruction to the control unit 1121.

  The output control unit 1133 connects, for example, an output device 1113 such as a display or a speaker, and outputs the processing result of the control unit 1121 to the output device 1113 as necessary.

  The program storage unit 1135 is configured by a ROM (Read Only Memory) or the like, and stores a program executed by the control unit 1121.

  The storage unit 1137 includes, for example, a storage device such as a hard disk device or a RAM (Random Access Memory), and temporarily stores the audio waveform data sent from the data recording unit 1139 and the audio waveform data after waveform expansion processing. . The storage unit 1137 sends the temporarily stored audio waveform data to the control unit 1121 or sends it to the data recording unit 1139 via the control unit 1121.

  The data recording unit 1139 is, for example, a CD-RW drive, and reads audio waveform data stored in the recording medium 1115 in accordance with an instruction from the control unit 1121. Further, the expanded audio waveform data is written in the recording medium 1115.

  The control unit 1121 performs waveform expansion processing on the audio waveform data temporarily stored in the storage unit 1137, and stores the expanded audio waveform data in the storage unit 1137. In the waveform expansion process, the control unit 1121 divides the audio waveform data into several divided waveforms in units of repetition. Then, a new segment waveform is generated at the boundary between the segment waveforms based on the segment waveforms before and after the boundary, and the segment waveform is inserted into the boundary to expand the voice waveform data. By repeating such an insertion operation, audio waveform data that ultimately becomes a specified multiple is generated. Since the waveform data is discrete data, when the designated multiple is M times, the decompressed speech waveform data has M times as many samples as the number of samples of the speech waveform data to be processed.

In the present embodiment, the original speech waveform is divided by the control unit 1121 using the repetition unit as a pitch. As a result, the original audio signal is treated as a series of _N pitch waveforms S ₀ , S ₁ ,..., S _N-1 as shown in FIG. Is called.

  As shown in FIG. 2, let the sampling time interval be q. For example, if the sampling frequency is 8 kHz as described above, q is 12.5 μs which is the reciprocal thereof.

Further, the pitch waveform S _j (0 ≦ j ≦ N−1) is expressed by pl (j) discrete data inside the audio processing device 1101. Therefore, pitch waveform S _j has time length pl (j) × q. Hereinafter, this may simply be described as the length of the pitch waveform being pl (j).

  Thus, although the sampling time interval q is constant, the length of each pitch waveform is generally not nearly constant although it is generally a close value. This is useful when the audio signal is expanded more than twice, as will be described in detail later.

The pitch waveform S _j (0 ≦ j ≦ N−1) is a series of wave heights of pl (j) sampling points {s _{j, 0} , s _{j, 1} ,..., S _{j, i} ,. , S _{j, pl (j) -1} }. This is expressed as S _j = {s _{j, 0} ,..., S _{j, pl (j) −1} } below.

For example, in FIG. 2 showing the time change of the audio signal, the i-th sampling point (white square mark in FIG. 2) of the j-th pitch waveform S _j is a time length (i−) from the start point of the pitch waveform S _j. 1) At the position of xq, the wave height is s _{j, i-1} . More specifically, for example, when the start point of the audio signal is taken as the origin of time, the wave height at time 7q when the first pitch waveform is continued is s ₀ , 7 (the eighth white circle mark from the left in FIG. 2). ).

  Waveform expansion is performed in pitch units. Basically, the expansion is performed by repeatedly interrupting some pitch waveform between two adjacent pitch waveforms. According to such repetition, in principle, the original speech waveform data can be expanded to an arbitrary length.

  Obviously, the pitch waveform to be interrupted preferably has some correlation with one or both of the two pitch waveforms to be interrupted. If an abrupt pitch waveform having no correlation with two interrupted pitch waveforms is interrupted, the sound signal is lost at that portion, and the sound quality deteriorates.

  To make the new pitch waveform to be interrupted correlate with one or both of the two existing pitch waveforms to be interrupted, for example, simply copy one of the two existing pitch waveforms May be a new pitch waveform.

  However, if it does in this way, as a result of interrupting a new pitch waveform, the completely same pitch waveform will be continued. In general, when human speech is to be decompressed, if decompression is performed by continuing identical waveforms continuously, the sound becomes unnatural.

  As described above, when stretching in units of pitch waveforms, it is not appropriate to insert a completely abrupt waveform between the two original pitch waveforms. It is also undesirable to insert such a simple copy. In other words, the new waveform to be inserted must not have a completely unrelated shape compared to the shape of the two pitch waveforms that were adjacent before the insertion, and the original two pitch waveforms Don't be too similar to one.

  In order to meet such a demand, as shown in FIG. 3, in the present embodiment, the original two pitch waveforms (FIG. 3A) having the insertion destination of the new pitch waveform as a boundary are appropriately subjected to waveform deformation processing. Then, a new pitch waveform having an intermediate shape between the two original pitch waveforms is generated by superimposing such processed waveforms (FIG. 3B), and the new pitch is generated. An audio signal is expanded by inserting a waveform (FIG. 3D).

  In the waveform deformation process (FIG. 3B), the new pitch waveform (FIG. 3C) generated as a result of subsequent superposition is intermediate between the two original pitch waveforms (FIG. 3A). Any processing can be used as long as it has a simple shape. A simple processing method using weighting will be described later.

  Even if new pitch waveforms are inserted one by one at all boundaries of the original N pitch waveforms, the degree of expansion is only doubled. If it is desired to extend more than twice, it is necessary to insert a plurality of new pitch waveforms at at least one boundary of the original pitch waveform boundary.

  Here, the above-mentioned request that the identical waveform should be avoided again becomes a problem. That is, when a plurality of new pitch waveforms are inserted at one of the boundaries of the original pitch waveform, the plurality of new pitch waveforms are not identical to any of the original pitch waveforms, However, it should not be the same for at least the adjacent relationship after the decompression process. This means that when a new waveform is generated from the original two pitch waveforms by the procedure shown in FIG. 3, a plurality of methods for generating the new waveform are required. Since only one generation method is required, the processing complexity increases.

  Therefore, in the following, a simpler process, i.e., a process when the expansion is smaller than twice will be described. Then, the process in the case of larger than 2 times is demonstrated.

  If the expansion is smaller than twice, new waveforms are inserted into some of the boundaries of the original pitch waveform, and nothing is inserted into the remaining some.

  Therefore, in order to make the audio signal reach the specified degree of expansion, it is necessary to distinguish between the boundary where the new pitch waveform is to be inserted and the boundary where the new pitch waveform is not inserted among the boundaries of the original pitch waveform.

  At this time, a priority order is assigned to each boundary of the original pitch waveform according to some criteria, and a new pitch waveform is inserted in order from a higher priority order to expand the audio signal, extending to the specified degree of expansion. The policy of stopping the insertion at the time is simple.

  Therefore, the next issue is how to determine the priority order. Qualitatively, it is obvious that the priority should be set higher at the boundary where two consecutive original pitch waveforms are similar. At this boundary, since the time variation of the pitch waveform is originally gentle, even if a new pitch waveform, which is an artificial waveform, is inserted, it is not so noticeable, that is, it is considered that the voice quality is hardly deteriorated. .

Then, the next problem is how to define an index that quantitatively indicates how similar the two pitch waveforms are. Any index may be used as long as it determines the priority for inserting a new pitch waveform. However, in this embodiment, the mean square error is adopted with emphasis on simplicity. This is called the degree of divergence. It can be said that the smaller the degree of divergence is, the two pitch waveforms forming the boundary are similar, that is, the time change of the pitch waveform is more gradual. Therefore, a new pitch waveform is inserted preferentially from a boundary with a small deviation. As shown in FIG. 2, the deviation degree e _j between the pitch waveforms S _j and S _{j + 1} is
e _j = {(s _{j, 0} −s _{j + 1,0} ) ² +
... + (s _{j, pl (j) -1} -s _{j + 1, pl (j) -1} ) ² } / pl (j)
Define as follows.

A specific procedure of audio signal expansion processing using the degree of divergence will be described with reference to FIG. It is assumed that the storage unit 1137 in FIG. 1 has already stored an audio signal to be processed in a state of being divided into _N pitch waveforms S ₀ ,. Each pitch waveform S _j (0 ≦ j ≦ N−1) is a sequence of wave heights {s _{j, 0} ,..., S _{j, pl (j) −1} } collected at the sampling time interval q.

First, the control unit 1121 secures an area in the storage unit 1137 for storing the structure S _magnified having the expanded pitch waveform as a component. As the initial value S _init of S _magnified, a value obtained by directly copying the sequence {S ₀ ,..., S _N-1 } of the pitch waveform of the original audio signal stored in the storage unit 1137 is employed. . That is, S _magnified = S _init = {S ₀ ,..., S _N-1 } (step S1401).

  As described above, the processing procedure differs depending on whether the expansion ratio is smaller or larger than two. Therefore, it is determined whether or not the magnification is smaller than 2 times (step S1403). If it is larger than 2 times (step S1403; No), an expansion process of 2 times or more which will be described in detail later is performed (step S1419). On the other hand, if it is determined that the magnification is smaller than 2 (step S1403; Yes), the process proceeds to step S1405 for determining an interrupt destination of a new waveform.

As described above, adjacent pitch waveform divergence e ₀ ,..., E _N is used to determine the interrupt destination of the new waveform. The degree of divergence is calculated in advance and stored in the storage unit 1137. Since a pitch waveform boundary with a smaller divergence degree is preferentially inserted into a new pitch waveform, the control unit 1121 stores a variable k indicating the order when the divergence degrees are arranged in ascending order as a counter in the register. The initial value is 1 (step S1405). Discrepancy e _0, · · ·, when rearranging the e _N in ascending order, for example, if e _j is the k-th, it is a boundary between the pitch waveforms S _j pitch waveforms S _{j + 1,} a new pitch This means that the waveform is inserted into the kth candidate.

Next, the control unit 1121 searches for the kth smallest one of the divergence degrees e ₀ ,..., E _N stored in the storage unit 1137, extracts the subscript, and stores k. A counter register different from the register is loaded (step 1407). For example, a small discrepancy in the k th if e _j, loading j. At this time, a sort operation for the degree of divergence is required. For this work, any known sort method such as bubble sort is used.

As described above, if j is loaded in step 1407, a new pitch waveform is inserted at the boundary between pitch waveform S _j and pitch waveform S _{j + 1} . The new pitch waveform is generated from the original two pitch waveforms as shown in FIG. Therefore, when the control unit 1121 generates a new pitch waveform based on the original two pitch waveforms in accordance with the program stored in the program storage unit 1135, first, the original pitch waveforms S _j and S _{j + 1} are obtained. You need to find out what it is. Therefore, the control unit 1121 loads the pitch waveforms S _j and S _{j + 1} from the storage unit 1137 to the general-purpose register (step S1409).

Next, the control unit 1121, the wave height column data s _j of pitch waveforms _{S j, 0, ···, s} j, pl (j) _-1, wave height column of pitch waveforms S _{j + 1} data s _{j + 1 , 0} ,..., S _{j + 1, pl (j + 1) −1} and new pitch waveform D _j = {d _{j, 0} ,..., D _{j, pl (j) −1} } Is generated according to the program stored in the program storage unit 1135 (S1411).

As described above, the new pitch waveform D _j must have an intermediate form between the two original pitch waveforms S _j and S _{j + 1} . On the other hand, any generation method may be used as long as an intermediate form is generated. In the present embodiment, a new waveform is generated by weighted addition. Weighted addition has the advantage of being easy to calculate.

Specifically, the original two pitch waveforms are weighted as indicated by the chain line in FIG. As shown in the left side of FIG. 3B, S _j that becomes a past pitch waveform in time when viewed from the new pitch waveform after expansion is multiplied by a linearly changing weighting coefficient starting from 0 and ending at 1. Waveform {s _{j, 0} × 0 / (pl (j) −1), s _{j, 1} × 1 / (pl (j) −1),..., S _{j, pl (j) −1} × ( pl (j) -1) / (pl (j) -1)}
Is generated. On the other hand, by multiplying S _{j + 1} , which becomes a pitch waveform of the future in time as viewed from the new pitch waveform after expansion, by a linearly changing weighting coefficient starting with 1 and ending with 0, FIG. Waveform as shown on the right side {s _{j + 1, 0} × (pl (j) −1) / (pl (j) −1), s _{j + 1, 1} × (pl (j) −2) / (pl (J) -1), ..., _{sj + 1, pl (j) -1} * 0 / (pl (j) -1)}
Is generated. Thereafter, the two weighted waveforms are superposed to form a new pitch waveform D _j = {d _{j, i} (0 ≦ i ≦ pl (j) −1) | d _{j, i} = (s _{j, i} × i + s _{j +1, i} * (pl (j) -1-i)) / (pl (j) -1)}
Is generated.

  The reason for using the mountain-shaped weighting coefficient as shown by the chain line in FIG. 3A can be caused by inserting an artificial waveform if a new pitch waveform is created with an emphasis on the shape of the waveform closer to the insertion location. This is because it can be expected to minimize the unnaturalness of speech.

When the generation of the new pitch waveform D _j is completed, it is inserted between the original pitch waveforms S _j and S _{j + 1} as shown in FIG. j) Decompress only. In order to perform this operation inside the speech processing apparatus 1101, the control unit 1121 extracts the pitch waveform sequence S _magnified from the storage unit 1137, _inserts D _j between S _j and S _{j + 1} , and _sets S _magnified . _{S magnified = {S 0, ···} , S j, D j, S j + 1, ···, S N-1} is updated as. Further, the control unit 1121 measures the length of the updated S _magnified for comparison with the expansion target value, and then stores the updated S _magnified in the storage unit 1137 (step S1413).

It is assumed that the target value for expansion is transmitted to the control unit 1121 via the input control unit 1131 using the input device 1111, for example, before the start of the expansion process. In step S1415, it is determined whether or not the length of S _magnified measured in step S1413 has reached the target expansion value. If the target value has been reached (step S1415; Yes), no further expansion is necessary, so the control unit 1121 terminates the expansion process, and the data recording unit 1139 _sets S _magnified at this time as the final expansion result. Recording is performed on the recording medium 1115, or the completion of the expansion process is output to the output device 1113 via the output control unit 1133 to notify the user.

On the other hand, when it is determined that the S _magnified updated in step S1413 has not reached the target length (step S1415; No), the control unit 1121 generates a new pitch waveform and adds it to the S _magnified. Since S _magnified should be expanded, the process returns to step S1407 to search for a boundary to which a new pitch waveform is to be inserted. At this time, in order to select a boundary having the highest insertion priority from boundaries where a new pitch waveform has not yet been inserted, the counter k representing the priority is incremented by 1 (step S1417).

Since the degree of expansion is smaller than twice (step S1403; Yes), S _magnified always reaches the target value (step _S1415 ; Yes) until the new pitch waveform is completely inserted into all the boundaries between the original pitch waveforms. ) And the process is complete.

  Next, the process (step S1419) when the designated expansion is larger than twice (step S1403; No) will be described again.

  FIG. 5 is a flowchart showing a specific procedure of the decompression process in the case where it is larger than twice. Compared with the case where the extension is smaller than twice, the processing becomes complicated as much as a plurality of new pitch waveforms must be inserted at one or more of the boundaries between the two original continuous pitch waveforms. Basically, a new pitch waveform is inserted in order from the boundary with the highest priority, as in the case where the extension is smaller than twice. However, the signal cannot be extended to the target length even after the new pitch waveform has been inserted one by one at every boundary. Therefore, it is necessary to insert new pitch waveforms one by two in turn, starting from the boundary with the highest priority again. If the target length is not reached even after the procedure is completed, another new pitch waveform is inserted three times this time. For this reason, the expansion process (FIG. 5) when it is larger than twice roughly counts how many times the entire boundary is circulated with respect to the expansion process (FIG. 4) when it is smaller than twice. This is a process with one more loop.

  The above-mentioned counter that counts how many times the entire boundary is circulated will be referred to as an all pitch waveform cycle number counter m. The initial value is set to m = 1 to represent the first round (step S1501).

Next, in order to assign a priority as an insertion destination of the new pitch waveform to the boundary formed by the original pitch waveform, a counter k representing the priority is initialized to k = 1 (step S1503), and the kth candidate is selected. seeking index j representing a certain boundary (step S1505), taken out S _j and S _{j + 1} is the original pitch waveforms corresponding to j (step S1507). These are the same as the procedures (steps S1405, S1407, and S1409 in FIG. 4) performed at the time of expansion of 2 times or less.

Thereafter, in step S1509, a new pitch waveform is generated from S _j and S _{j + 1} . When the cycle between all the pitch waveforms is the first time, that is, when m = 1, only one new pitch waveform needs to be generated. Therefore, a new waveform generation method in the expansion procedure of twice or less (step S1411 in FIG. 4). There is no change. As described above, one new pitch waveform may be generated by performing an appropriate weighting and adding.

  However, the maximum number of cycles that can be extended by performing only one cycle between all pitch waveforms is up to twice. In the case of the expansion process of 2 times or more, the circulation between all pitch waveforms is not completed once. That is, the maximum value of the total pitch waveform cycle count counter m is always 2 or more. In the m-th cycle between all pitch waveforms, a boundary having a high priority determined by the small degree of divergence is sequentially selected from the boundaries where (m−1) new pitch waveforms have already been inserted. The (m−1) new pitch waveforms that have been inserted into the boundary are discarded once, and instead, m new pitch waveforms are newly generated and inserted into the boundary. As a result, the audio signal can be expanded by one pitch waveform.

  The problem here is that the new pitch waveform must be different from both of the two original pitch waveforms, and if a plurality of new pitch waveforms are generated, at least after the insertion, they are adjacent. The things that fit together must be different. As described above, this is a request for avoiding a situation in which a human voice is unnatural due to repetition of the same waveform.

  Of these problems, the request that must be different from any of the two original pitch waveforms is the same request as when the degree of expansion is less than twice. Therefore, a new pitch waveform may be generated by weighted addition as in the case where the degree of expansion is smaller than twice.

  On the other hand, a request that a plurality of new pitch waveforms to be newly generated must be different from each other occurs only when the degree of expansion is twice or more. In order to respond to such a request, for example, by preparing various types of methods for generating a new pitch waveform from the original two pitch waveforms, and adopting a new generation method every time a new pitch waveform is generated, It is conceivable to avoid overlapping a plurality of new pitch waveform shapes.

  However, it is cumbersome to prepare multiple types of generation methods in advance, and when the degree of expansion is large, all of the prepared types of generation methods are used up, and the same new pitch waveform generated once is generated. There is a possibility that the pitch waveform must be generated again, and the above-described request may not be met.

  Therefore, in this embodiment, a simpler and more reliable method is adopted. As described above, the length of each original pitch waveform is generally the same as the order but is expected to be different. In particular, the possibility that the adjacent pitch waveforms have the same length is almost zero. This is because a human voice is mainly assumed as a processing target of the voice processing apparatus 1101 according to the present embodiment, and it is difficult to think that a human voice remains in a steady state for a long time.

  In the present embodiment, such a property that the lengths of the original adjacent pitch waveforms are different is used. That is, as a method for generating a new pitch waveform, only one type of the above-described method of performing weighted addition of the original two pitch waveforms is used. This simplifies the process. In order to satisfy the requirement that there is a difference between the new pitch waveforms, the generated new pitch waveform is expanded and contracted in the time axis direction so that the same new pitch waveform does not occur. .

  Expansion and contraction of the new pitch waveform in the time axis direction means changing the length of the waveform. Therefore, the next issue is the range of changes. Obviously, in order to minimize degradation of the quality of the voice after decompression, it is appropriate to select the required number of lengths within a range that falls between the lengths of the original two pitch waveforms.

Also in this embodiment, a plurality of new pitch waveforms are generated by such a method. That is, when generating the m new pitch waveform (step S1509), first, when stretched 2 times or less to produce a new pitch waveform D _j by the same weighted addition (Fig. 3, step S1411 of FIG. 4). By expanding and contracting D _j on the time axis, m variations of the pitch waveform D _{1, j} = {d _{1, j, 0} ,..., D _{1, j, pl (j) −1} }, ..., _{Dm, j} = {dm _{, j, 0} , ..., dm _{, j, pl (j + 1) -1} }
Is generated. That is, paying attention to the fact that the length pl (j) of the original two adjacent pitch waveforms S _{j and} the length pl (j _{+ 1} ) of S _{j + 1} are almost certainly different as described above, D _{1, j} , ..., the length of D _{m, j} is adjusted to m lengths from pl (j) to pl (j + 1). As for how to divide between pl (j) and pl (j + 1) into m stages, in the present embodiment, it is assumed that the division is most simply and proportionally. That is, the length of the i-th (1 ≦ i ≦ m) new pitch waveform D _{i, j} is
pl (j) + [{pl (j + 1) −pl (j)} × (i−1) / (m−1)]
And

  The length of a plurality of new pitch waveforms to be inserted is gradually increased or decreased from the length of the past pitch waveform in the original pitch waveform to the length of the pitch waveform in the future in time. In addition to simplifying the process, it is considered that there is an advantage of minimizing degradation of the quality of the voice after expansion because the change of the voice signal due to the insertion is gentle.

  There are various methods for changing the length of the pitch waveform, but it is easy to change the sampling position. This method will be described in detail later.

Eventually, m new pitch waveforms D _{i, j} (1 ≦ i ≦ m) are generated from the original two pitch waveforms S _j and S _{j + 1} (step S1509). The length of the D _{1, j} is likewise pl (j) and _{S j,} D _m, the length of _j is also pl and _{S j + 1 (j + 1} ), D 2, j ~D m-1, The length of _j is between pl (j) and pl (j + 1).

When m new pitch waveforms are generated in step S1509, these are added to S _magnified which is the result of voice expansion at this time, and S _magnified is updated. S _magnified is a candidate for the extension signal prepared during the extension process (step S1401 in FIG. 4). More specifically, the control unit 1121 takes out the pitch waveform sequence S _magnified from the storage unit, and _inserts D _{1, j} ,..., D _{m, j} between S _j and S _{j + 1.} by, S _Magnified = updated as _{{S 0, ···, S j} , D 1, j, ···, D m, j, S j + 1, ···, S N-1}, Save in the storage unit 1137. At the same time, the length of S _magnified is measured in order to be useful for comparison with the target length (step _S1511 ).

  As described above, when m new pitch waveforms are generated and inserted at an appropriate position in the m-th cycle between all pitch waveforms, they are inserted at the positions in the previous pitch waveform cycle (m−1). The new waveforms are discarded. On the other hand, the (m−1) or m new pitch waveforms already inserted at the boundary that is not the insertion target at this time are left as they are.

Then, since the inserted pitch waveform is increased by one, the audio signal S _magnified is expanded by one pitch waveform length. Therefore, it is determined whether or not the length of S _magnified measured in step _S1511 has reached the target value of the degree of expansion at this point (step S1513). If it has reached (step S1513; Yes) ), The expansion process is completed, and S _magnified is _set as the final expansion result.

  If the target value has not been reached (step S1513; No), it is next determined whether k, which is a priority counter, is smaller than the number N of the original pitch waveforms. If it is smaller (step S1515; Yes), there are boundaries where m new waveforms have not yet been inserted, so k is incremented by 1 (step S1519), and a process of searching for the next priority insertion position is searched. Return to (step S1505).

  On the other hand, when k becomes N (step S1515; No), since m new waveforms have been inserted into all the boundaries, m is increased by 1 (step S1517), and all pitch waveforms are interleaved. Are repeated in order of priority determined by the degree of divergence (step S1503).

In principle, S _magnified can be lengthened by repeating the cycle between all pitch waveforms, that is, by increasing m. Therefore, according to the above procedure, the process always reaches the designated degree of expansion without falling into an infinite loop, and the process is completed.

(Embodiment 2)
Embodiment 2 according to the present invention will be described below with reference to the drawings. The sound processing device according to the second embodiment is a sound reduction device. The configuration is almost the same as that of the audio processing apparatus according to the first embodiment shown in FIG. 1, but an audio reduction process is performed instead of the audio expansion process.

  In the waveform reduction process, the control unit 1121 divides the original speech waveform data to be processed into pitch waveform units. Then, based on the two consecutive pitch waveforms, a new pitch waveform is generated, and the original two consecutive pitch waveforms are replaced with the new pitch waveform, thereby increasing the length of the pitch waveform one by one. Reduce voice waveform data. By repeating this replacement operation, speech waveform data having a specified length is finally generated.

  In order to reduce the waveform, it is easiest to thin out the pitch waveform appropriately. However, if the thinning is simply performed, the difference between the pitch waveforms adjacent to each other as a result of the thinning may be too large. There is. Then, an effect similar to that in which a discontinuous portion occurs in the audio signal is produced, and as a result, there is a possibility that the quality of the audio signal after the reduction is deteriorated.

  Since it is decided to perform replacement rather than simple decimation, it is desirable that the newly generated pitch waveform for replacement has some correlation with one or both of the two pitch waveforms to be replaced. Is clear. If an abrupt pitch waveform having no correlation with the two pitch waveforms to be removed is arranged, the audio signal is not smooth at that portion and the sound quality is deteriorated.

  In order for the new pitch waveform generated for replacement to have a correlation with one or both of the two existing pitch waveforms to be replaced, for example, which of the two existing pitch waveforms A simple copy of these may be used as a new pitch waveform.

  However, if it does in this way, as a result of interrupting a new pitch waveform, the completely same pitch waveform will be continued. Generally, when exactly the same waveform continues, unnatural sound is produced.

  As described above, when reducing in units of pitch waveforms, it is not appropriate to arrange a completely abrupt waveform instead of the original two pitch waveforms. It is also undesirable to place a simple copy of it. In other words, the new pitch waveform to be used for replacement must not have an irrelevant shape compared to the shape of the two pitch waveforms that were adjacent before the replacement, and the original two pitch waveforms. Don't be too similar to one of the waveforms.

  In order to meet such a demand, as shown in FIG. 6, in the present embodiment, the original two pitch waveforms (FIG. 6A) existing at the planned placement destination of the new pitch waveform are appropriately deformed. Processing (FIG. 6 (b)) and superimposing such processed waveforms to generate a new pitch waveform having an intermediate shape between the two original pitch waveforms, The audio signal is reduced by arranging it instead of the pitch waveform (FIG. 6C).

  In the above-described waveform deformation process (FIG. 6B), the new pitch waveform (FIG. 6C) generated as a result of subsequent superposition is intermediate between the two original pitch waveforms (FIG. 6A). Any processing can be used as long as it has a simple shape. A simple processing method using weighting will be described later.

  Even if new pitch waveforms are arranged one by one at all boundaries of the original N pitch waveforms, the reduction degree is only 0.5 times. A problem arises when it is desired to reduce the length to be shorter than 0.5 times.

  Such a problem is related to the problem in the first embodiment in which special consideration is required for the extension processing of twice or more.

  However, unlike the enlargement process, the reduction process is a process that only reduces the number of pitch waveforms. Therefore, there is no need to consider that a plurality of generated new pitch waveforms that are problematic in the first embodiment must be different from each other.

Therefore, the process of reducing the image to be shorter than 0.5 times performed in the present embodiment is simpler than the process of reducing the image to be longer than 2 times performed in the first embodiment. That is, when it is specified that the reduction is shorter than 0.5 times, the reduced audio signal immediately after the reduction to 0.5 times is considered as the new initial value of the audio signal, and the same Repeating the procedure is enough. In principle, 0.25 times if it is repeated twice, (0.5) ⁱ times if it is repeated ⁱ times, and so on, unless it is required to reduce the pitch waveform length to the length. can do.

  Therefore, only the procedure that can reduce the original audio signal to 0.5 times will be described below.

  When the required reduction degree is larger than 0.5 times, an original pitch waveform that is replaced with a new waveform and a waveform that remains as it is are generated.

  Therefore, in order to make the audio signal reach the specified degree of reduction, it is necessary to discriminate between the original pitch waveform to be replaced with the new pitch waveform and the one to be left as it is.

Therefore, as in the case of the first embodiment, the divergence degree e _j (0 ≦ j ≦ N−1) is calculated for each boundary of the original pitch waveform, and priority is given to the two pitch waveforms sandwiching the boundary where this value is small. Therefore, it is replaced with a new pitch waveform. Then, a policy is adopted in which the audio signal is reduced in order from the place with the highest priority, and the replacement is stopped when the audio signal is reduced to a specified reduction degree.

  It can be said that the time change of the audio signal was originally gentler as the divergence degree was smaller. Therefore, it is expected that the quality of the audio signal is less likely to deteriorate in such a portion even if an artificial operation of replacement is applied. From this, it can be said that it is reasonable to determine the priority order of replacement locations based on the degree of deviation.

A specific procedure of the audio signal reduction process in the present embodiment will be described with reference to FIG. It is assumed that the storage unit 1137 in FIG. 1 has already stored an audio signal to be processed in a state of being divided into _N pitch waveforms S ₀ ,. Each pitch waveform S _j (0 ≦ j ≦ N−1) is a sequence of wave heights {s _{j, 0} ,..., S _{j, pl (j) −1} } collected at the sampling time interval q.

First, the control unit 1121 reserves an area in the storage unit 1137 for storing the structure S _reduced having the expanded pitch waveform as a component. As the initial value S _init of S _reduced, a value obtained by directly copying the sequence {S ₀ ,..., S _N-1 } of the pitch waveform of the original audio signal stored in the storage unit 1137 is employed. . That is, S _reduced = S _init = {S ₀ ,..., S _N-1 } (step S1701).

As described above, adjacent pitch waveform divergence e ₀ ,..., E _N is used to determine the placement location of the new waveform. The divergence degree is calculated in advance and stored in the storage unit 1137. Since a pitch waveform boundary with a smaller divergence degree is preferentially inserted into a new pitch waveform, the control unit 1121 stores a variable k indicating the order when the divergence degrees are arranged in ascending order as a counter in the register. The initial value is 1 (step S1703). Discrepancy e _0, · · ·, when rearranging the e _N in ascending order, if e _j is the k-th, it boundary between the pitch waveforms S _j pitch waveforms S _{j + 1} is the replacement operation It means to become the kth candidate as a target.

Next, the control unit 1121 searches for the kth smallest one of the divergence degrees e ₀ ,..., E _N stored in the storage unit 1137, extracts the subscript, and stores k. A counter register different from the register is loaded (step 1705). For example, a small discrepancy in the k th if e _j, loading j.

As described above, if j is loaded in step 1705, the pitch waveform S _j and the pitch waveform S _{j + 1} are replaced with the new pitch waveform. The new pitch waveform is generated from the original two pitch waveforms as shown in FIG. Therefore, when the control unit 1121 generates a new pitch waveform based on the original two pitch waveforms in accordance with the program stored in the program storage unit 1135, first, the original pitch waveforms S _j and S _{j + 1} are obtained. You need to find out what it is. Therefore, the control unit 1121 loads the pitch waveforms S _j and S _{j + 1} from the storage unit 1137 to the general-purpose register (step S1707).

Next, the control unit 1121, the wave height column data s _j of pitch waveforms _{S j, 0, ···, s} j, pl (j) _-1, wave height column of pitch waveforms S _{j + 1} data s _{j + 1 , 0} ,..., S _{j + 1, pl (j + 1) −1} and new pitch waveform D _j = {d _{j, 0} ,..., D _{j, pl (j) −1} } Is generated according to the program stored in the program storage unit 1135 (S1709).

  In the present embodiment, a new waveform is generated by weighted addition. Weighted addition has the advantage of being easy to calculate.

Specifically, the original two pitch waveforms are weighted as shown by the chain line in FIG. S _j is multiplied by a linearly changing weighting coefficient starting with 1 and ending with 0, thereby giving a waveform {s _{j, 0} × (pl (j) −1) as shown on the left side of FIG. / (Pl (j) -1), _{sj, 1} * (pl (j) -2) / (pl (j) -1), ..., _{sj, pl (j) -1} * 0 / ( pl (j) -1)}
Is generated. On the other hand, S _{j + 1} is multiplied by a linearly changing weighting coefficient starting with 0 and ending with 1, thereby obtaining a waveform {s _{j + 1, 0} × 0 / (Pl (j) -1), _{sj + 1, 1} * 1 / (pl (j) -1), ..., _{sj + 1, pl (j) -1} * (pl (j) -1 ) / (Pl (j) -1)}
Is generated. Thereafter, the two weighted waveforms are superposed to form a new pitch waveform C _j = {c _{j, i} (0 ≦ i ≦ pl (j) −1) | c _{j, i} = {s _{j, i} × (pl (J) -1-i) + _{sj + 1, i} * i) / (pl (j) -1)}
Is generated.

The reason why the valley-shaped weighting coefficient as shown by the chain line in FIG. 6A is used is that the pitch waveform S _j-1 that is left without being replaced among the pitch waveforms S _j and S _{j + 1 that} are removed by the replacement. If a new pitch waveform is created with a greater emphasis on the shape of the closer portion and the portion closer to S _{j + 2} , it can be expected to minimize the unnaturalness of speech that can occur by inserting an artificial waveform. It is.

When the generation of the new pitch waveform C _j is completed, as shown in FIG. 6C, it is arranged in place of the original pitch waveforms S _j and S _{j + 1} so that the length of the audio signal is pl ( Shrink by j + 1). In order to perform this operation inside the voice processing device 1101, the control unit 1121 takes out the pitch waveform sequence S _reduced = {S _{r, 0} ,..., S _{r, Nk} } from the storage unit 1137. In addition, since S _reduced has undergone (k-1) substitutions, the number of pitch waveforms is reduced by (k-1), and is eventually composed of a total of (N-k + 1) pitch waveforms. . The control unit 1121 deletes S _{r, j + 1} from S _reduced to thereby obtain a first provisional pitch waveform sequence S _tmp = {S _{r, 0} ,..., S _{r, j} , S _{r, j + 2} , _.. , S _{r, Nk} } is generated, and then an operation of substituting C _j into S _{r, j} is performed. Subsequently, S _r, the operation of substituting the value of _{j + 2} S _r, the _{j + 1, S r, j} + 3 of the value S _r, the operation of substituting the _{j + 2,} etc., S _{r, Nk} This is continued until the value of is substituted for S _{r and Nk−1} . Thus, the second provisional pitch waveform sequence S ′ _tmp = {S _{r, 0} ,..., S _{r, Nk−1} } is generated, and S _reduced is updated to S _reduced = S ′ _tmp . Further, the control unit 1121 measures the length of the updated S _reduced for comparison with the reduction target value, and then stores the updated S _reduced in the storage unit 1137 (step S1711).

It is assumed that the reduction target value is transmitted to the control unit 1121 via the input control unit 1131 using the input device 1111, for example, before the reduction process is started. In step S1713, it is determined whether or not the length of S _reduced measured in step S1711 has reached the target value for reduction. If the target value has been reached (step S1713; Yes), no further reduction is necessary, so the control unit 1121 terminates the expansion process, and the data recording unit 1139 uses S _reduced as the final reduction result. Recording is performed on the recording medium 1115, and the fact that the reduction process is completed is output to the output device 1113 via the output control unit 1133 to notify the user.

On the other hand, when it is determined that the updated S _reduced in step S1711 has not reached the target length (step S1713; No), the control unit 1121 generates a new pitch waveform and performs a replacement operation to perform the S _reduced. Therefore, the process returns to step S1705 to search for a boundary that is a new pitch waveform placement destination. At this time, in order to select the boundary having the highest insertion priority from the boundaries not yet related to the arrangement of the new pitch waveform, the counter k indicating the priority is incremented by 1 (step S1715).

As described above, since the specified degree of reduction is assumed to be a value larger than 0.5, S _reduced is always set to the target value until the replacement operation is performed for all the boundaries formed by the original pitch waveforms. It reaches (step S1713; Yes), and the processing is completed.

  As described above, even when the specified reduction degree is smaller than 0.5, the process can be reduced to the target length by repeating the process shown in FIG.

(Embodiment 3)
An apparatus according to the third embodiment of the present invention includes an apparatus that has both the voice expansion function of the apparatus according to the first embodiment and the voice reduction function of the apparatus according to the second embodiment. That is, the device according to the present embodiment is an audio expansion / contraction device.

  The configuration is the audio processing apparatus 1101 shown in FIG. 1 as in the apparatuses according to the first and second embodiments. However, the program storage unit 1135 stores a program that can handle both audio expansion and audio reduction processes.

  The apparatus according to the present embodiment first determines whether to expand or contract, and performs the same operation as the apparatus according to the first embodiment if it is to be expanded, and if it is to be reduced, the apparatus according to the second embodiment. It is assumed that the same operation is performed.

  The user designates the expansion magnification or the reduction magnification M (M> 0) using the input device 1111 shown in FIG. Then, according to the program stored in program storage unit 1135, control unit 1121 first determines whether M is 0 <M <1, M> 1, or M = 1. To do. Next, when it is determined that 0 <M <1, the control unit 1121 determines that the reduction process is requested, and causes the apparatus according to the present embodiment to operate as the apparatus according to the second embodiment. If it is determined that M> 1, the same decompression process as that performed by the apparatus according to the first embodiment is executed. Note that when M = 1, it means equal magnification, and the processing result is a copy of the original speech waveform data to be processed as it is.

(Example of how to change the length of the pitch waveform)
In the apparatus according to the first embodiment, a process for variously changing the length of the pitch waveform is required when the expansion is twice or more. As a method for changing the pitch waveform length, as described above, it is easy to change the sampling position. This method will be specifically described below with reference to FIG.

FIG. 8A schematically shows the time dependence of the pitch waveform S _j . The horizontal axis represents time. If the sampling period is q, the pitch waveform S _j is expressed as an array of wave heights for each time interval q in the speech processing apparatus 1101 in FIG. 1 (white circles in FIG. 8A), and the pitch waveform length is pl. (J) × q.

  The principle of the pitch waveform length changing method by changing the sampling position is the same whether the pitch waveform length is extended or reduced. Here, the case of reduction will be described as an example.

Assume that the sampling period is r (where r> q), and re-sampling is performed. The time at which this new sampling is performed is indicated by a black triangle mark in FIG. 8A, and the waveform acquired by the new sampling is indicated by a white square mark in FIG. 8A. In order to facilitate understanding, it is further assumed that the relationship of r to q is as shown in FIG. That is, when the time series is followed in order from the start time of the pitch waveform S _j , the new sampling time is gradually delayed from the original sampling time, and at the time when the pitch waveform S _j ends, the old and new samplings are performed. Suppose that the times match.

  The new sampling period r is originally longer than the original sampling period q. However, in the calculation, the wave height sampled at the new sampling period r is acquired at the original sampling period q. Treat as there is. That is, from the state where the pitch waveform having the pitch waveform length pl (j) × q is discretely represented by the wave height array of the new sampling period r (FIG. 8B), the wave height array is left as it is. Then, processing is performed assuming that the sampling period is q. This processing means that the sampling points represented by white squares in FIG. 8B are moved as indicated by white arrows.

As a result of the movement, a new pitch waveform shown in FIG. 8C was generated. As is clear from this figure, the new pitch waveform is the original pitch waveform S _j shortened on the time axis. The waveform length pl ′ of the new pitch waveform is shorter than the waveform length p (j) × q of the original pitch waveform S _j by time q. That is,
pl ′ = {pl (j) −1} × q
It is. Alternatively, the magnification is [{pl (j) -1} / {pl (j)}]
It can also be said that double reduction has been completed.

The above description has been made on the assumption that resampling is performed faithfully to the original pitch waveform. However, depending on the operation of the audio processing device 1101, it may be considered that the pitch waveform S _j as analog data is discarded after the pitch waveform S _j is once sampled at the period q. Alternatively, it may be possible to omit the procedure required for resampling even if it is not discarded.

  In such a case, re-sampling at a position where there is no sample is necessary. For this purpose, various methods are conceivable, but the simplest method is a method using linear interpolation described below.

In FIG. 8D, a pitch waveform as analog data is drawn with a chain line, and a portion where the first sampling is performed is represented by a white circle. When re-sampling is attempted at the time indicated by the black triangle in FIG. 8D for waveform reduction processing, the pitch waveform as analog data has already been lost, or analog data processing has been repeated. I don't want to. At this time, exact resampling is no longer possible except for a time that happens to overlap the initial sampling time. Therefore, as a second best measure, as shown in FIG. 8D, for example, a re-sampling value at the time when the original sampling interval is internally divided into t: u between the original wave heights a and b is necessary. Linear interpolation a + [{(b−a) × t} / (t + u)}
Approximate with the value calculated by.

In addition, this invention is not limited to the said embodiment, A various deformation | transformation and application are possible.
For example, in the first embodiment, when the audio signal is expanded more than twice, a plurality of new pitch waveform lengths in which the pitch waveform length is changed little by little are generated. On the other hand, even in the case of an extension process that is longer than twice, a process that is similar to the process that is performed when it is requested to reduce the length to half or less in the second embodiment may be used. In other words, when the expansion of 2 times or more is requested, once the expansion is performed up to 2 times, the expansion result is regarded as the original audio data again, and the expansion processing up to 2 times is performed again. Without adjusting the pitch waveform length, it is possible to expand up to four times when viewed from the true original audio data. If this is repeated, in principle, it can be expanded to an arbitrary magnification.

  The voice processing device 1101 may further include a communication control unit that communicates with other devices via a communication network such as the Internet, and the expanded and contracted voice waveform data is transmitted to the other devices via the communication control unit. You may make it transmit to. In addition, voice waveform data may be received from another device via this communication control unit, and expanded or contracted.

  Alternatively, the output device 1113 may be capable of outputting sound, and the sound signal obtained as a result of the expansion / contraction process may be immediately reproduced so that the user can judge the performance on the spot.

  Note that the information processing apparatus for realizing the audio processing apparatus 1101 according to the embodiment of the present invention can be realized using a normal computer system, not necessarily a dedicated system. For example, by storing and distributing a general-purpose computer in a computer-readable recording medium (FD, CD-ROM, DVD, etc.) storing a program for executing the above-described operation, and installing the program in the computer The voice processing device 1101 that executes the above-described processing can be configured. In addition, the program may be stored in a disk device included in a server device on a communication network such as the Internet and downloaded to a computer, for example.

  Further, when the OS shares a part of the above-described processing or when the OS constitutes a part of the constituent elements of the present invention, a program excluding the part is stored in the recording medium. It may be distributed or downloaded to a computer. Also in this case, the recording medium stores a program for executing each function or each step executed by the computer.

1 is a block diagram of a speech processing apparatus according to Embodiment 1 of the present invention. It is a figure which shows the state by which the audio | voice waveform which is a process target is time-divided into the pitch waveform. It is a figure which shows a mode that a new pitch waveform is produced | generated based on two continuous pitch waveforms, and it is inserted. It is a flowchart for demonstrating the process when the expansion | extension mainly smaller than 2 times is requested | required among the audio | voice expansion | extension processes which concern on Embodiment 1 of this invention. It is a flowchart for demonstrating a process when expansion | extension larger than 2 times is requested | required among the audio | voice expansion | extension processes which concern on Embodiment 1 of this invention. It is a figure which shows a mode that a new pitch waveform is produced | generated based on two continuous pitch waveforms, and the original two pitch waveforms are substituted by it. It is a flowchart for demonstrating the audio | voice reduction | decrease process which concerns on Embodiment 2 of this invention. It is a figure which shows a mode that the length of a pitch waveform is changed by the change of a sampling point, and primary interpolation.

Explanation of symbols

1101 ... Audio processing device, 1111 ... Input device, 1113 ... Output device, 1115 ... Recording medium, 1121 ... Control unit, 1131, ... Input control unit, 1133 ... Output control 1135 ... Program storage unit 1137 ... Storage unit 1139 ... Data recording unit

Claims (16)

A storage unit for storing a speech waveform that is a time series of segmented waveforms;
With respect to two consecutive segment waveforms in the speech waveform stored in the storage unit, a weighting factor that increases linearly in time with respect to the front segment waveform and time with respect to the rear segment waveform. Each using a weighting factor that decreases linearly in each case, and a new segmented waveform generating means for generating one new segmented waveform by weighted addition,
A plurality of insertions each having a different length between the length of the front segment waveform and the length of the rear segment waveform by expanding / contracting the new segment waveform generated by the new segment waveform generation means on the time axis Insertion segment waveform generation means for generating segment waveform,
Insertion means for inserting a plurality of insertion segment waveforms generated by the insertion segment waveform generation unit between the front segment waveform and the rear segment waveform;
A voice waveform output means for outputting a voice waveform expanded by inserting an insertion section waveform by the insertion means;
A voice decompression device comprising: Similarity calculation means for calculating a similarity indicating how similar the two continuous waveform segments are;
Target length arrival determining means for determining whether or not the length of the voice waveform expanded by inserting the insertion section waveform by the inserting means has reached a predetermined voice waveform target length;
Further comprising
The insertion means inserts the insertion segment waveform in descending order of similarity between the two consecutive segment waveforms based on the similarity calculated by the similarity calculation unit,
The target length arrival determination means determines whether or not the length of the voice waveform expanded every time the insertion section waveform is inserted by the insertion means has reached a predetermined voice waveform target length,
When it is determined that the target length has been achieved by the target length arrival determining means, the insertion by the inserting means is terminated, and when it is determined that the target length has not been achieved, the insertion means continues to insert according to the order. Do,
The audio decompression apparatus according to claim 1, wherein: The insertion segment waveform generation means is configured to insert the insertion segment waveform between the two consecutive segment waveforms included in the speech waveform stored in the storage unit by the insertion means. If the target length arrival determining means determines that the target length has not been reached, increase the number of segmented waveforms to be generated,
The speech decompression apparatus according to claim 2, wherein The insertion segment waveform generation means generates a plurality of insertion segment waveforms each having a different length by changing the sampling position of the new segment waveform.
The audio expansion device according to any one of claims 1 to 3, wherein   5. The speech decompression apparatus according to claim 1, further comprising speech segmentation means for segmenting a speech waveform to be processed into time series of segmented waveforms. The voice dividing means detects a pitch break of the voice, and generates a time series of segmented waveforms by division according to the break;
The voice decompression apparatus according to claim 5. An audio expansion method for expanding an audio waveform that is a time series of segmented waveforms,
For two consecutive segment waveforms in the speech waveform, a weighting factor that increases linearly in time for the preceding segment waveform and decreases linearly in time for the rear segment waveform. A new segment waveform generation step for generating one new segment waveform by weighting and adding the weighting factors to be used, respectively,
A plurality of insertions having different lengths between the length of the front segment waveform and the length of the rear segment waveform by expanding and contracting the new segment waveform generated in the new segment waveform generation step on the time axis. An insertion segment waveform generation step for generating a segment waveform; and
An insertion step of inserting a plurality of insertion segment waveforms generated in the insertion segment waveform generation step between the front segment waveform and the rear segment waveform;
An audio decompression method consisting of: A similarity calculation step for calculating a similarity indicating how similar the two continuous waveform segments are;
A target length attainment determining step for determining whether or not the length of the voice waveform expanded by inserting the insertion segment waveform by the insertion step has reached a predetermined voice waveform target length;
Further comprising
The insertion step inserts an insertion segment waveform in descending order of similarity between the two consecutive segment waveforms based on the similarity calculated by the similarity calculation step;
The target length arrival determination step determines whether or not the length of the voice waveform expanded every time the insertion segment waveform is inserted by the insertion step has reached a predetermined voice waveform target length,
When it is determined that the target length has been achieved by the target length arrival determination step, the insertion by the insertion step is terminated, and when it is determined that the target length has not been achieved, the insertion step continues to insert according to the order. Do,
The audio decompression method according to claim 7. Said insertion segment waveform generation step, the insertion be performed is inserted in the insertion segment waveform by step, the target length reach determination steps for between two segment waveform successive all included in the sound voice waveform If it is determined that the target length has not been reached, increase the number of segmented waveforms to be generated.
The audio decompression method according to claim 8. The insertion segment waveform generation step generates a plurality of insertion segment waveforms each having a different length by changing the sampling position of the new segment waveform.
The audio decompression method according to claim 7, wherein the audio decompression method is performed.   The speech decompression method according to any one of claims 7 to 10, further comprising a speech division step of dividing a speech waveform to be processed into time series of segmented waveforms. A computer of a voice expansion device that expands a voice waveform that is a time series of segmented waveforms stored in a storage unit,
With respect to two consecutive segment waveforms in the speech waveform stored in the storage unit, a weighting factor that increases linearly in time with respect to the front segment waveform and time with respect to the rear segment waveform. Each newly using a weighting factor that decreases linearly, and a new segmented waveform generating means for generating one new segmented waveform by weighted addition,
A plurality of insertions each having a different length between the length of the front segment waveform and the length of the rear segment waveform by expanding / contracting the new segment waveform generated by the new segment waveform generation means on the time axis Insertion section waveform generation means for generating section waveforms,
Insertion means for inserting a plurality of insertion segment waveforms generated by the insertion segment waveform generation unit between the front segment waveform and the rear segment waveform;
A voice waveform output means for outputting a voice waveform expanded by inserting an insertion section waveform by the insertion means;
Program to function as. Computer
Similarity calculation means for calculating a similarity indicating how similar the two continuous waveform segments are;
Target length arrival determining means for determining whether or not the length of the voice waveform expanded by inserting the insertion section waveform by the inserting means has reached a predetermined voice waveform target length;
Further function as
The insertion means inserts the insertion segment waveform in descending order of similarity between the two consecutive segment waveforms based on the similarity calculated by the similarity calculation unit,
The target length arrival determination means determines whether or not the length of the voice waveform expanded every time the insertion section waveform is inserted by the insertion means has reached a predetermined voice waveform target length,
When it is determined that the target length has been achieved by the target length arrival determining means, the insertion by the inserting means is terminated, and when it is determined that the target length has not been achieved, the insertion means continues to insert according to the order. Do,
The program according to claim 12, wherein the program is made to function as follows. The insertion segment waveform generation means is configured to insert the insertion segment waveform between the two consecutive segment waveforms included in the speech waveform stored in the storage unit by the insertion means. If the target length arrival determining means determines that the target length has not been reached, increase the number of segmented waveforms to be generated,
The program according to claim 13, wherein the program is made to function as follows. The insertion segment waveform generation means generates a plurality of insertion segment waveforms each having a different length by changing the sampling position of the new segment waveform.
The program according to any one of claims 12 to 14, wherein the program is made to function as follows. Computer
The program according to any one of claims 12 to 15, further functioning as voice dividing means for dividing a voice waveform to be processed into a time series of segmented waveforms.

Images