JP2008209447A - Time-axis expansion and compression method, time-axis expansion and compression device, program and basic cycle specifying method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To reduce a processing amount for specifying a basic cycle. <P>SOLUTION: A first processing section 31 calculates a similarity degree R of waveforms of adjoining sections S1 and S2 in an input signal VIN, having the same section length L, for a plurality of section lengths L changed by a change width D1 corresponding to two or more samples. A second processing section 32 calculates the similarity degree R of waveforms of sections S1 and S2 of the input signal VIN, for the plurality of section lengths L generated by changing a temporary section length LX in which the similarity degree calculated by the first processing section 31 indicates similarity, by a change width D2 that is smaller than the change width D1. An expansion and compression section 24 generates an output signal VOUT of a waveform in which the input signal VIN is expanded or compressed on a time-axis, so that the section length L in which the similarity degree R calculated by the first processing section 31 and the second processing section 32 shows similarity, is a basic cycle LW. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a technique for specifying a time length (hereinafter referred to as “basic period”) of a section in which similar waveforms repeat among input signals such as audio signals.

Conventionally, a technique for compressing or expanding an input signal in the direction of the time axis while maintaining the pitch and formant by deleting or inserting a waveform corresponding to a basic period has been proposed (for example, Patent Document 1). In the technique of Patent Document 1, the process of calculating the waveform similarity for the first section and the second section that are adjacent to each other with the same section length in the input signal has one section length for the first section and the second section. The section length having the highest degree of similarity is specified as the basic period by being repeated a plurality of times while changing by the number of samples.
Japanese Patent No. 3465628

  The technique of Patent Document 1 attempts to reduce the processing amount by calculating the similarity after thinning samples for each of the first section and the second section. However, there are cases where the amount of processing is not sufficiently reduced by simply thinning out samples for which the similarity is calculated, and further reduction of the amount of processing is desired. In view of the above circumstances, an object of the present invention is to solve the problem of reducing the processing amount for specifying the basic period.

  In order to solve the above-described problem, the time axis companding method according to the present invention includes a first interval and a first interval that are adjacent to each other in a sample sequence in which a plurality of samples of an input signal are arranged in time series. A first stage (for example, the process of FIG. 3) in which a process of calculating the similarity of waveforms for two sections is performed for a plurality of section lengths changed by a first change width corresponding to two or more samples, and a sample The processing for calculating the similarity of the waveform for the first section and the second section of the column is the second in which the section length (temporary section length) in which the similarity calculated in the first stage shows similarity is smaller than the first change width. The second stage (for example, the process of FIG. 4) executed for a plurality of section lengths changed by the change width, and the section length in which the similarity calculated in the first stage and the second stage shows similarity is set as a basic period. Waveform of input signal expanded or compressed on time axis And a companding process generating a force signal.

  In the above configuration, the waveform similarity between the first section and the second section is similar after the waveform similarity between the first section and the second section is calculated while changing the section length by the first change width. Similarities are calculated for a plurality of section lengths in which a specific section length is changed by a second change width smaller than the first change width. Therefore, it is possible to specify the basic period with high accuracy compared to the method of calculating the similarity only for the section length in which the section length is changed by the first change width, and for all the changes in the second change width. The amount of processing can be reduced as compared with the configuration for calculating the similarity for the section length.

  In the second stage of the time axis companding method according to a preferred aspect of the present invention, a plurality of (for example, N pieces in FIGS. 3 and 4) selected in descending order of similarity calculated in the first stage. For each of the section lengths, the similarity between the first section and the second section is calculated for a plurality of section lengths obtained by changing the section length by the second change width. According to the above aspect, compared with the case where only one section length calculated in the first stage is the target of processing in the second stage, the specific accuracy of the basic period can be increased.

  The second change width is arbitrarily set in a range that satisfies the condition that it is smaller than the first change width. However, if the second change width is set to a size corresponding to the interval between successive samples in the sample sequence of the input signal, the effect of specifying the basic period with high accuracy becomes particularly significant.

  In the first stage of the time axis companding method according to a specific aspect of the present invention, the variable relating to the first stage is variably controlled according to the sampling frequency of the input signal. For example, the first change width increases as the sampling frequency of the input signal is higher. According to the above aspect, it is possible to suppress the processing amount in the first stage even when the sampling frequency is high.

  In the time axis companding method according to a specific aspect of the present invention, the section length for calculating the similarity in the second stage is smaller as the distribution width of the similarity calculated for each section length in the first stage is smaller. (For example, the re-search width LD in FIG. 4) is reduced. There is a high possibility that a clear periodicity does not appear in a section in which the similarity distribution width is small (for example, a section of silent or unvoiced consonant) in the input signal. Therefore, according to the above aspect, the processing in the second stage is performed while maintaining the specific accuracy of the fundamental period at a high level for the voiced sound section of the input signal (that is, the section in which the fundamental period should be originally identified). It is possible to reduce the amount. Further, if the distribution of similarity calculated for each section length in the first stage is below a predetermined value, the processing amount in the second stage will be reduced if the similarity calculation in the second stage is stopped. The effect becomes even more remarkable.

  The present invention is also specified as a device that compresses or expands an input signal in the direction of the time axis. The time-axis companding device according to the present invention calculates the similarity of waveforms for a first section and a second section that are equal in section length and are adjacent to each other in a sample sequence in which a plurality of samples of an input signal are arranged in time series. First processing means for executing the processing to be performed for a plurality of section lengths that are changed by the first change width corresponding to two or more samples, and the waveform similarity for the first section and the second section of the sample sequence A second processing means for executing a calculation process for a plurality of section lengths obtained by changing the section length in which the degree of similarity calculated by the first processing means is similar by a second change width smaller than the first change width; Companding processing means for generating an output signal having a waveform in which the input signal is expanded or compressed on the time axis with the interval length in which the similarity calculated by the first processing means and the second processing means is similar as the basic period. . Also with the above apparatus, the same operation and effect as the time-axis companding method according to the present invention are combined.

  The time axis companding device according to the present invention is realized by hardware (electronic circuit) such as a DSP (Digital Signal Processor) dedicated to specific signal processing, or a general-purpose operation such as a CPU (Central Processing Unit). This is also realized by cooperation between the processing device and the program. The program according to the present invention performs a process of calculating a waveform similarity for a first section and a second section that are equal in section length and are adjacent to each other in a sample sequence in which a plurality of samples of an input signal are arranged in time series. First processing means for executing a plurality of section lengths that are changed by a first change width corresponding to two or more samples, and a process for calculating a waveform similarity for the first section and the second section of the sample sequence Second processing means for executing a plurality of section lengths obtained by changing the section length in which the degree of similarity calculated by the first processing means is similar for each second change width smaller than the first change width, and the first processing means And a function that causes the computer to function as a companding processing means for generating an output signal having a waveform in which the input signal is expanded or compressed on the time axis, with the section length indicating the similarity calculated by the second processing means as a basic period. Ah . By the above program, the same operations and effects as the time-axis companding device of the present invention are exhibited. The program of the present invention is provided to a user in a form stored in a portable recording medium such as a CD-ROM and installed in a computer, or provided from a server device in a form of distribution via a network. Installed on the computer.

  One embodiment of the present invention is a method for specifying a basic period of an input signal based on a sample sequence in which a plurality of samples of the input signal are arranged in time series. In the basic period specifying method according to the present embodiment, the processing for calculating the waveform similarity for the first section and the second section having the same section length and adjacent to each other in the sample sequence corresponds to two or more samples. The first stage executed for a plurality of section lengths changed by one change width and the process of calculating the waveform similarity for the first section and the second section of the sample sequence are similarities calculated in the first stage. The second stage executed for a plurality of section lengths obtained by changing the section length indicating similarity by a second change width smaller than the first change width, and the similarities calculated in the first stage and the second stage are similar. And a third stage that specifies the indicated section length as a basic period. According to the above method, it is possible to specify the basic period with high accuracy while reducing the processing amount. Note that the present invention can also be specified as a device (basic cycle specifying device) or program for realizing the basic cycle specifying method according to the above aspect.

<A: First Embodiment>
FIG. 1 is a block diagram showing a configuration of a time axis companding device according to the first embodiment of the present invention. The time axis companding apparatus 100 is an apparatus that generates an output signal VOUT having a waveform obtained by extending or compressing an input signal (audio signal) VIN indicating a waveform of a musical tone or a voice in the direction of the time axis. As shown in FIG. 1, the time axis companding device 100 includes a storage circuit 10, a control circuit 20, and an output circuit 40.

  A sample string of the input signal VIN is supplied to the memory circuit 10 from the outside. The sample string is a set in which a large number of samples extracted from the input signal VIN are arranged in time series. The storage circuit 10 is an input buffer circuit that sequentially stores each sample.

  The control circuit 20 is an arithmetic processing unit (CPU) that generates a sample sequence of the output signal VOUT from the sample sequence of the input signal VIN by executing a program. The output circuit 40 is an output buffer circuit that sequentially accumulates each sample of the output signal VOUT and sequentially outputs each sample at a predetermined period.

  As illustrated in FIG. 1, the control circuit 20 functions as a basic period specifying unit 22 and a companding processing unit 24 by executing a program. However, the basic period specifying unit 22 and the companding processing unit 24 may be realized by a hardware circuit such as a DSP. The basic period specifying unit 22 specifies a basic period LW that is a time length of a section in which similar waveforms of the input signal VIN repeat. As shown in FIG. 1, the basic period specifying unit 22 is divided into a first processing unit 31 and a second processing unit 32.

  FIG. 2 is a conceptual diagram for explaining an outline of operations of the first processing unit 31 and the second processing unit 32. In part (a) of FIG. 2, each sample P of the input signal VIN is shown by a white circle and a black circle. Part (b) of FIG. 2 is a conceptual diagram showing the operation of the first processing unit 31, and part (c) of FIG. 2 is a conceptual diagram showing the operation of the second processing unit 32.

  As shown in part (b) and part (c) of FIG. 2, each of the first processing unit 31 and the second processing unit 32 has waveforms in the sections S1 and S2 having the same section length L in the input signal VIN. The similarity R is calculated a plurality of times while changing each section length L (the number of samples P belonging to each of the sections S1 and S2). The section S1 and the section S2 are adjacent to each other on the time axis. Each time the first processing unit 31 and the second processing unit 32 calculate the similarity R, the similarity R and the section length L at the time of calculation are associated with each other and stored in the storage circuit 10. It is possible to employ any known method for calculating the similarity R. However, in this embodiment, the similarity R increases as the waveform is more similar (higher similarity) between the sections S1 and S2. Assume that the similarity R is calculated so as to decrease. For example, the similarity R is a square error between each sample P of the waveform in the section S1 and each sample P of the waveform in the section S2.

  As shown in part (b) of FIG. 2, the first processing unit 31 uses the section S1 and the section for each of the plurality of section lengths L changed by the change width D1 within the range from the maximum value LMAX to the minimum value LMIN. The similarity R with S2 is calculated. The change width D1 is a numerical value of 2 or more. In part (a) of FIG. 2, the sample P after thinning out a number of samples P of the input signal VIN with an argument (“4” in the present embodiment) for a predetermined period is shown as a white circle for convenience, and the rest Sample P (thinned sample P) is shown as a black circle. As shown in part (b) of FIG. 2, the change width D1 is the interval between the samples P after the thinning of the input signal VIN (that is, the interval DS of the samples P that precede and follow the input signal VIN before the thinning is 2). Equivalent to more than one). The change width D1 in FIG. 2 corresponds to four intervals DS. In other words, the first processing unit 31 sets the section length L of all types (“LMAX−LMIN + 1” types) changed by “1” corresponding to one sample P within the range from the maximum value LMAX to the minimum value LMIN. Is calculated for the remaining section length L when thinned out with the argument (D1) for a predetermined period, and the calculation of the similarity R is omitted for the section length L thinned out.

  Among all the section lengths L for which the first processing unit 31 has calculated the similarity degree R, a predetermined number of section lengths L (hereinafter referred to as the order in which the waveforms are similar between the sections S1 and S2). In particular, “temporary section length LX” is selected as a target of processing by the second processing unit 32. In part (c) of FIG. 2, one section length L (LMAX-2 · D1) having a small similarity R calculated by the first processing unit 31 is representatively shown as a provisional section length LX.

  As shown in part (c) of FIG. 2, the second processing unit 32 determines a section for each of a predetermined number ((2 · LD)) of section lengths L obtained by increasing or decreasing each provisional section length LX by the change width D2. The similarity R between S1 and section S2 is calculated. The change width D2 is smaller than the change width D1. The change width D2 in this embodiment is “1” corresponding to one interval DS (one sample).

  The numerical value LD (hereinafter referred to as “re-search width LD”) defines the number of section lengths L for calculating the similarity R by changing the change width D2 from one temporary section length LX. Part (c) of FIG. 2 illustrates the case where the re-search width LD is set to “3”. Therefore, as shown in part (c) of FIG. 2, the second processing unit 32 determines the interval from the interval length “LX-3” for each provisional interval length LX extracted from the result of the processing by the first processing unit 31. The similarity R is calculated for the section lengths L up to the length “LX + 3” (excluding the section length LX for which the similarity R has already been calculated).

  The section length L (the section length L having the most similar waveform in the sections S1 and S2) having the minimum similarity R calculated by the first processing section 31 and the second processing section 32 is specified as the basic period LW. The companding processing unit 24 generates a sample sequence of the output signal VOUT based on the basic period LW specified by the basic period specifying unit 22 and the companding rate C specified from the outside. The companding rate C is set in accordance with, for example, a user operation on an input device (not shown).

  When the companding ratio C indicates compression of the input signal VIN, the companding processing unit 24 synthesizes (for example, crossfade) the section S1 and the section S2 of the basic period LW in the input signal VIN, and generates a combined waveform. The output signal VOUT is generated by replacing the section S1 and the section S2. When the companding rate C indicates the expansion of the input signal VIN, the companding processing unit 24 synthesizes the section S1 and the section S2 of the basic period LW in the input signal VIN, and the synthesized waveform is represented by the sections S1 and S2. The output signal VOUT is generated by inserting it between the section S2. It should be noted that any known method can be employed for companding the input signal VIN in accordance with the basic period LW and the companding rate C.

  Next, a specific example of an operation in which the first processing unit 31 specifies N (N is a natural number) provisional section lengths LX will be described with reference to FIG. The first processing unit 31 initializes the variables LX [1] to LX [N] and variables RX [1] to RX [N] stored in the storage circuit 10 to zero (step SA1). The variables LX [1] to LX [N] at the stage where the processing of FIG. 3 is completed correspond to the N provisional section lengths LX. The variables RX [1] to RX [N] are the similarity R between the section S1 and the section S2 when each of the variables LX [1] to LX [N] is the section length L.

  The first processing unit 31 sets the section length L of the sections S1 and S2 to the maximum value LMAX (step SA2), and calculates the waveform similarity R for the sections S1 and S2 of the section length L (step SA3). ). The waveform of the section S1 is expressed by L samples P from the time point TC corresponding to the unprocessed oldest sample P to the time point "TC + L-1", and the waveform of the section S2 is expressed from the time point "TC + L" to the time point "TC + 2L". -1 "is represented by L samples P.

  The first processing unit 31 stores the similarity R calculated in step SA3 in the variable RX [1] and the variable RP, and stores the current section length L (maximum value LMAX) in the variable LX [1] and the variable LP. (Step SA4). Further, the first processing unit 31 subtracts the change width D1 from the section length L (step SA5). As shown in FIG. 2, the change width D1 is a numerical value of 2 or more. Then, the first processing unit 31 calculates the similarity R for the section S1 and the section S2 with the section length L after the update in step SA5 by the same method as in step SA3 (step SA6).

  The first processing unit 31 initializes the variable i to “1” (step SA7). Next, the first processing unit 31 determines whether or not the similarity R calculated in the immediately preceding step SA6 is less than the variable RX [i] (step SA8). When the result of step SA8 is affirmative (when the similarity of the waveform is high between the section S1 and the section S2 having the section length L), the first processing unit 31 sets the variable LX stored in the storage circuit 10 at this stage. [i] to LX [N-1] are shifted to variables LX [i + 1] to LX [N] and variables RX [i] to RX [N-1] are changed to variables RX [i + 1] to LX [ N] (step SA9). The first processing unit 31 updates the variable RX [i] to the current similarity R and updates the variable LX [i] to the current section length L, and then proceeds to step SA11 (step S11). SA10). On the other hand, if the result of step SA8 is negative, the first processing unit 31 shifts the process to step SA11 without executing the processes of step SA9 and step SA10.

  In step SA11, the first processing unit 31 adds “1” to the variable i, and determines whether or not the variable i after the addition exceeds the numerical value N (step SA12). If the result of step SA12 is negative, the first processing unit 31 repeats the processing from step SA8 to step SA11 for the updated variable i. On the other hand, when the result of step SA12 is affirmative, the first processing unit 31 proceeds to step SA13. As described above, the processing from step SA6 to step SA12 is repeated, so that the variable LX [i] has a waveform in order of decreasing similarity R calculated by the first processing unit 31 (waveforms in the sections S1 and S2). The i-th section length L is stored in the order of the similarity of the sections S1 and S2 and the similarity R when the sections S1 and S2 are set to the section length LX [i] is stored in the variable RX [i]. Stored.

  In step SA13, the first processing unit 31 sets a value obtained by subtracting the change width D1 from the current section length L as a new section length L. Then, the first processing unit 31 determines whether or not the updated section length L in step SA13 is less than the minimum value LMIN (step SA14). If the result of step SA14 is negative, the first processing unit 31 repeats the processing from step SA6 to step SA12 for the updated section length L. When the result of step SA14 is affirmative (that is, when the processing is completed for all the section lengths L changed by the change width D1 in the range from the maximum value LMAX to the minimum value LMIN), the first processing unit 31 performs FIG. Terminate the process.

  As described above, the processing from step SA6 to step SA13 is repeated while changing the section length L by the change width D1 within the range from the maximum value LMAX to the minimum value LMIN. At the stage where the processing of FIG. 3 is completed, N section lengths L are counted as the provisional section length LX as a temporary section length LX in the order of decreasing similarity R (in order of increasing waveform similarity between section S1 and section S2). 1] to LX [N], and the similarity R corresponding to each of the N provisional section lengths LX is stored in variables RX [1] to RX [N].

Next, a specific example of an operation in which the second processing unit 32 specifies the basic period LW will be described with reference to FIG.
The second processing unit 32 initializes the variable i to “1” (step SB1), and determines whether the variable RX [i] is less than the variable RP (the initial value is set in step SA4). (Step SB2). If the result of step SB2 is affirmative, the second processing unit 32 updates the variable LP to the variable LX [i] and updates the variable RP to the variable RX [i] (step SB3). On the other hand, if the result of step SB2 is negative, the process of step SB3 is not executed. That is, the variable LP is maintained at the provisional section length LX [i] at which the similarity RX [i] is the smallest at the present time, and the variable RP is maintained at the similarity RX [i].

  Next, the second processing unit 32 updates the section length L to the added value of the variable LP and the re-search width LD (step SB4). The second processing unit 32 calculates the similarity R for the section S1 and the section S2 with the updated section length L in step SB4 by the same method as in step SA3 and step SA6 (step SB5).

  The second processing unit 32 determines whether or not the similarity R calculated in step SB5 is less than the variable RP (step SB6). If the result of step SB6 is affirmative, the second processing unit 32 stores the similarity R calculated in step SB5 in the variable RP and stores the current section length L in the variable LP (step SB7). On the other hand, if the result of step SB6 is negative, the process of step SB7 is not executed.

  Next, the second processing unit 32 sets a numerical value obtained by subtracting the change width D2 from the current section length L as a new section length L (step SB8). As shown in FIG. 2, the change width D2 is smaller than the change width D1 in step SA5 or step SA13. Since the similarity R of the provisional section length LX [i] has already been calculated, in step SB8, the second processing unit 32 determines that the section length L after subtraction of the change width D2 is the provisional section length LX [i]. If it matches, the change width D2 is further subtracted. That is, the process from step SB5 to step SB7 is omitted for the provisional section length LX [i].

  The second processing unit 32 determines whether or not the updated section length L in step SB8 is equal to or greater than the difference value (LP−LD) between the variable LP and the re-search width LD (step SB9). If the result of step SB9 is affirmative, the second processing unit 32 proceeds to step SB5. As described above, in steps SB5 to SB9, the similarity R is determined for a predetermined number (“2 · LD”) of section lengths L that are increased or decreased by the change width D2 with the provisional section length LX [i] as the median. While being calculated, the variable LP and the variable RP are updated according to the degree of similarity R.

  If the result of step SB9 is negative (that is, the processing from step SB5 to step SB9 is completed for “2.LD” section lengths L with the i-th provisional section length LX [i] as the median value) ), The second processing unit 32 adds “1” to the variable i (step SB10), and determines whether or not the variable i after the addition exceeds the numerical value N (step SB11). If the result of step SB11 is negative, the second processing unit 32 executes the process after step SB2 on the variable i after the update at step SB10. On the other hand, when the result of step SB11 is affirmative (that is, when the processing from step SB2 to step SB10 is completed for N temporary section lengths LX [1] to LX [N]), the second processing unit 32 After specifying the variable LP at the current stage as the basic period LW (step SB12), the processing in FIG. 4 is terminated.

  As described above, in this embodiment, first, N temporary section lengths LX are obtained by calculating the similarity R between the sections S1 and S2 while changing the section length L by the change width D1. Secondly, the basic period LW is specified by calculating the similarity R for “2 · LD” section lengths L in which the provisional section length LX is changed by a change width D2 smaller than the change width D1. The Therefore, the basic period LW is specified in comparison with the configuration in which the similarity R is calculated by changing the section length L by “1” over the entire range from the minimum value LMIN to the maximum value LMAX (hereinafter referred to as “comparative”). There is an advantage that the amount of processing to do is reduced.

  FIG. 5 is a chart in which the number of processes for calculating the similarity R is compared with this embodiment for each sampling frequency FS of the input signal VIN. In the figure, as shown in FIG. 2, the change width D1 is “4”, the change width D2 is “1”, the number N of the provisional section lengths LX specified by the first processing unit 31 is “8”, It is assumed that the search width LD is “3”. In addition, the ratio of the number of times of similarity calculation in the present embodiment to the number of times of calculation of similarity R in contrast is described as “ratio”.

  As shown in FIG. 5, when the sampling frequency FS of the input signal VIN is 16 kHz, the number of processes for calculating the similarity R is reduced to 44.8% in the present embodiment. The effect when compared with the proportionality becomes more prominent as the sampling frequency FS increases. For example, when the sampling frequency FS is 48 kHz, according to the present embodiment, the number of times of calculating the similarity R is reduced to 31.6% which is proportional.

  Further, since the second processing unit 32 calculates the similarity R for each section length L in which the provisional section length LX is changed by a change width D2 smaller than the change width D1, the section length L changed by the change width D1. Even if the first processing unit 31 calculates the similarity R only, the accuracy of specifying the basic period LW does not unduly decrease compared to the proportionality. Further details are as follows.

  6 to 8 are graphs showing the relationship between the section length L and the similarity R of the sections S1 and S2 set in the input signal VIN. FIG. 6 illustrates a case where the input signal VIN indicates voiced sound. As shown in the figure, for voiced sound, the waveform similarity R between the section S1 and the section S2 changes periodically with respect to the section length L, so the change width D1 is larger than the period of variation of the similarity R. If it is smaller, even if the section length L is changed by the change width D1 as in the present embodiment, the similarity R is the smallest (most) in the temporary section length LX specified by the first processing unit 31 in FIG. The provisional section length LX close to the section length L is included. Therefore, the second processing unit 32 calculates the similarity R for each section length L obtained by changing the provisional section length LX by the change width D2, so that the section length L (that is, the similarity R is the minimum in FIG. It is possible to specify the basic period LW) with an accuracy equivalent to the proportionality.

  FIG. 7 shows a case where the input signal VIN is recorded in a state where there is no voiced sound (hereinafter referred to as “silent state”), and FIG. 8 shows a case where the input signal VIN indicates an unvoiced consonant. In a silent state or a state in which a voiceless consonant exists, the similarity R varies irregularly with respect to the section length L. Therefore, the fundamental cycle LW that is actually specified differs between the present embodiment in which the first processing unit 31 changes the section length L by the change width D1 and the proportionality in which the section length L is changed by “1”. However, since the input signal VIN in a silent state or a state in which there is an unvoiced consonant is inherently poor in periodicity (a clear basic period LW does not appear), it is fundamental to change the section length L by the change width D1. Even if the period LW is different from proportional, it hardly affects the sound quality of the output signal VOUT. As described above, according to this embodiment, there is an advantage that the processing amount can be reduced while maintaining the accuracy of specifying the basic period LW to be equal to the proportionality.

<B: Modification>
Various modifications can be made to each of the above embodiments. An example of a specific modification is as follows. In addition, you may combine each following aspect suitably.

(1) Modification 1
In the above embodiment, the case where the change width D1 is fixed is exemplified, but the change width D1 (an argument between sample columns) may be variably controlled. For example, as shown in FIG. 9, a configuration is adopted in which the change width D1 is changed according to the sampling frequency FS of the input signal VIN. The frequency specifying unit 26 in FIG. 9 is means for specifying the sampling frequency FS of the input signal VIN based on, for example, a user operation on an input device (not shown) or an instruction from the host device. The first processing unit 31 controls the change width D1 so that the change width D1 increases as the sampling frequency FS specified by the frequency specifying unit 26 increases.

  When the change width D1 is a fixed value, the number of times the similarity R is calculated by the first processing unit 31 increases as the sampling frequency FS increases. According to the configuration of FIG. 9, since the change width D1 (an argument between samples P) increases as the sampling frequency FS increases, even if the sampling frequency FS is high, the specific accuracy of the basic period LW is reduced. It is possible to reduce the number of times of calculation of the similarity R by the first processing unit 31 without any problem. Note that numerical values other than the change width D1 may be changed according to the sampling frequency FS. For example, a configuration in which the first processing unit 31 controls the number N of the provisional section lengths LX according to the sampling frequency FS is also employed. For example, the number N is increased as the sampling frequency FS is higher.

(2) Modification 2
Clear periodicity appears in voiced sounds. Therefore, as understood from FIGS. 6 to 8, when the input signal VIN indicates voiced sound, the distribution width of the similarity R with respect to the section length L (the difference between the maximum value RMAX and the minimum value RMIN of the similarity R) (Value) ΔR is sufficiently large as compared with the case where the input signal VIN indicates a silent state or a case where the input signal VIN indicates a voiceless consonant (that is, when no periodicity appears in the input signal VIN). In other words, when the distribution width ΔR of the similarity R is large, the input signal VIN indicates voiced sound, and when the distribution width ΔR of the similarity R is small, the input signal VIN indicates a silent state or unvoiced consonant. There is. On the other hand, since no clear periodicity appears in the input signal VIN indicating silence or voiceless consonant, it is not meaningful to search for the fundamental period LW strictly.

  Therefore, when the distribution width ΔR of the similarity R calculated for the plurality of section lengths L by the first processing unit 31 is small (that is, when the input signal VIN is highly likely to indicate silence or voiceless consonant), the second processing unit A configuration for reducing the amount of processing by 32 is preferably employed. For example, first, as the distribution width ΔR of the similarity R is smaller, the re-search width LD (that is, the number of times the second processing unit 32 calculates the similarity R) used in step SB4 and step SB9 is the second process. A configuration in which the portion 32 is reduced is adopted. Second, a configuration in which the second processing unit 32 stops the processing of FIG. 4 (particularly the calculation of the similarity R) when the distribution width ΔR is less than a predetermined threshold is also employed. When the process of FIG. 4 is stopped, for example, the second processing unit 32 designates a predetermined basic period LW to the companding processing unit 24 regardless of the input signal VIN.

  According to the above configuration, it is possible to reduce the processing amount of the second processing unit 32 when the input signal VIN indicates silence or a voiceless consonant. Although the processing of the second processing unit 32 is simplified or omitted, when the input signal VIN is a voiced sound, the basic period LW is specified in the same manner as in the above form, so that the output signal As for VOUT, it is possible to maintain the sound quality equivalent to the above-mentioned form.

(3) Modification 3
Each numerical value used in the processing of FIG. 3 and FIG. 4 is appropriately changed. For example, according to the configuration in which the re-search width LD is set to a numerical value larger than the change width D1, the basic period LW can be specified with higher accuracy. When the re-search width LD is larger than the change width D1, the second processing unit 32 calculates the similarity R within the range (the range from “LX + LD · D2” to “LX−LD · D2”). The section length L for which the first processing unit 31 has already calculated the similarity R is included. Therefore, for the section length L for which the similarity R has already been calculated, a configuration in which the second processing unit 32 omits the processing from step SB5 to step SB8 is also employed.

(4) Modification
The first processing unit 31 stores the similarity R of all the section lengths L changed by the change width D1 in the storage circuit 10, and the second processing unit 32 performs the order of the similarity R in the storage circuit 10 in descending order. A configuration is also adopted in which the processing of FIG. 4 is executed with the N section lengths L selected as the temporary section length LX. That is, the configuration in which the N temporary section lengths LX are narrowed down in the processing by the first processing unit 31 (FIG. 3) is not necessarily required.

(5) Modification 5
The basic period specifying unit 22 is used in addition to the time axis companding device 100. For example, in a pitch detection device that outputs the reciprocal of the basic period LW as a pitch, the basic period specifying unit 22 may be used to specify the basic period LW from the input signal VIN.

It is a block diagram which shows the structure of the time-axis companding apparatus which concerns on embodiment of this invention. It is a conceptual diagram for demonstrating operation | movement of a 1st process part and a 2nd process part. It is a flowchart which shows operation | movement of a 1st process part. It is a flowchart which shows operation | movement of a 2nd process part. It is a chart for demonstrating the effect of this invention. It is a graph which shows the relationship between the section length and similarity when the input signal VIN shows a voiced sound. It is a graph which shows the relationship between the section length in case the input signal VIN shows silence, and a similarity. It is a graph which shows the relationship between the section length and similarity when the input signal VIN shows a voiced consonant. It is a block diagram which shows the structure of the time-axis companding apparatus which concerns on a modification.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 ... Time-axis companding device, 10 ... Memory circuit, 20 ... Control circuit, 22 ... Fundamental period specifying part, 24 ... Companding processing part, 31 ... First processing part, 32 ... Second Processing unit 40... Output circuit, VIN... Input signal, VOUT.

Claims (9)

The process of calculating the similarity of the waveforms for the first and second sections having the same section length and adjacent to each other in the sample sequence in which a plurality of samples of the input signal are arranged in time series is made into two or more samples. A first step of executing a plurality of section lengths changed by corresponding first change widths;
The process of calculating the waveform similarity for the first section and the second section of the sample sequence is the second change in which the section length indicating the similarity calculated in the first stage is smaller than the first change width. A second stage executed for a plurality of section lengths varied by width;
A companding process step of generating an output signal having a waveform obtained by expanding or compressing the input signal on a time axis, with the section length indicating the similarity calculated in the first step and the second step being similar. And a time axis companding method. In the second stage, for each of a plurality of section lengths selected in order of decreasing similarity calculated in the first stage, a plurality of sections in which the section length is changed by the second change width The time axis companding method according to claim 1, wherein a similarity between the first section and the second section is calculated for the length. The time-axis companding method according to claim 1, wherein the second change width corresponds to an interval between samples that follow each other in the sample row. The time axis companding method according to claim 3, wherein, in the first stage, the first change width is increased as the sampling frequency of the input signal is higher. 5. The number of section lengths for calculating similarity in the second stage is decreased as the distribution width of the similarity calculated for each section length in the first stage is smaller. 5. The time axis companding method described in 1. 5. The calculation of the similarity is stopped in the second stage when the distribution width of the similarity calculated for each section length in the first stage is less than a predetermined value. 5. Time axis companding method. The process of calculating the similarity of the waveforms for the first and second sections having the same section length and adjacent to each other in the sample sequence in which a plurality of samples of the input signal are arranged in time series is made into two or more samples. First processing means for executing a plurality of section lengths changed by corresponding first change widths;
The processing for calculating the similarity of the waveform for the first section and the second section of the sample sequence is performed by setting a second section length that is smaller than the first change width to a section length in which the similarity calculated by the first processing means shows similarity. A second processing means for executing a plurality of section lengths changed by a change width;
A companding processing means for generating an output signal having a waveform obtained by extending or compressing the input signal on the time axis, with a section length showing similarity between the calculated similarities of the first processing means and the second processing means as a basic period A time axis companding device comprising: Computer
The process of calculating the similarity of the waveforms for the first and second sections having the same section length and adjacent to each other in the sample sequence in which a plurality of samples of the input signal are arranged in time series is made into two or more samples. First processing means for executing a plurality of section lengths changed by corresponding first change widths;
The processing for calculating the similarity of the waveform for the first section and the second section of the sample sequence is performed by setting a second section length that is smaller than the first change width to a section length in which the similarity calculated by the first processing means shows similarity. A second processing means for executing a plurality of section lengths changed by a change width;
A companding processing means for generating an output signal having a waveform obtained by extending or compressing the input signal on the time axis, with a section length showing similarity between the calculated similarities of the first processing means and the second processing means as a basic period A program that functions as and. A method of identifying a basic period of an input signal based on a sample sequence in which a plurality of samples of the input signal are arranged in time series,
A plurality of processings in which the processing for calculating the similarity between the waveforms of the first section and the second section having the same section length in the sample sequence and being adjacent to each other is changed by a first change width corresponding to two or more samples. A first stage to perform on the section length;
The process of calculating the waveform similarity for the first section and the second section of the sample sequence is the second change in which the section length indicating the similarity calculated in the first stage is smaller than the first change width. A second stage executed for a plurality of section lengths varied by width;
And a third step of specifying, as a basic cycle, a section length in which the similarity calculated in the first step and the second step shows similarity.

Images