JPH0573089A - Speech reproducing method - Google Patents

Abstract

PURPOSE:To obtain a reproduced speech which can be heard as a natural speech even when its reproducing speed is varied and also reducible in process distortion in a voice section by finding the ratio of speech section length of specific block length and overall block length, comparing the value with a desired time compression ratio, and selecting a compressing method according to the result. CONSTITUTION:The ratio Rv of the total time length of the speech section in one block of an input signal and the time length of the whole block is found and compared with the compression ratio Rc of the time compression of the input signal. When Rc>=kRv (k: constant), only voiceless sections are compressed. When Rc<kRc, on the other hand, independent compression ratios are set for the voiceless sections and speech sections and the time compression is performed. Thus, an elongation ratio is varied for the voiceless sections and speech section according to the value Rv, so the speech which is close to the natural speech can be reproduced.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention, when reproducing a recorded conversational voice signal or the like, does not cause a change in the pitch even if the speed is changed to a speed other than the normal speed, and reproduces the sound without any inconvenience. It is about the method.

[0002]

2. Description of the Related Art Conventionally, various devices having a function of varying the reproduction speed of a recorded audio signal have been proposed. The simplest examples are cues (fast forward playback) and reviews (rewind playback) of analog tape recorders, but these high-speed playback voices have both pitch and speed changes compared to normal-speed voices. It is almost impossible to hear the contents. Also, a voice reproduction method utilizing a speed conversion process for compensating the pitch so as not to change even during high speed / low speed reproduction has been proposed. Hereinafter, a conventional audio reproduction method for performing speed conversion processing will be described with reference to the drawings.

FIG. 4 shows a flowchart of a conventional audio reproducing method. First, the time companding ratio R _CE is determined. Here, the time companding ratio R _CE is shown as follows.

[0004]

[Equation 1]

As is clear from (Equation 1), R _CE takes a value smaller than 1 when reproducing at a speed higher than normal speed, and R _CE is larger than 1 when reproducing at a speed lower than normal speed. Takes a value. The speed conversion process is performed on the input voice signal using this R _CE as a parameter. The speed conversion process is a process of changing only the speed while keeping the pitch as in the normal reproduction. For the speed conversion processing, see, for example, "Digital Speech Processing", Sadahiro Furui, TDHS algorithm of Tokai University Press. Then, in the case where the subsequent processing is repeatedly performed, it is confirmed whether R _CE is changed, R _CE is determined, and then the speed conversion processing is performed. By repeating the above operation, the audio signal is reproduced at the reproduction speed desired by the user.

[0006]

However, in the above configuration, the reproduction speed is changed by performing the speed conversion processing for uniformly expanding and contracting the time axis regardless of the state of the input signal. As a result, the pitch of the reproduced sound is preserved, but even when the reproduction speed is greatly changed from the normal speed, the time is expanded and contracted uniformly without distinction between the silent section and the voice section. Has a problem of feeling different unnatural reproduced sounds. Further, processing distortion due to speed conversion processing is always added to the voice section transmitting information.

The present invention solves the above problems, and a voice reproduction method for obtaining a reproduced sound which can be heard as natural as possible even if the reproduction speed is changed and the processing distortion of the voice section is reduced. The purpose is to provide.

[0008]

In order to achieve this object, a voice reproducing method according to the present invention considers a ratio R _V between a voice section length and a block length in one block, and considers a silent section and a voice section. A method is employed in which the compression ratio (expansion ratio) is set separately and time compression (time expansion) is performed. Also, the time compression ratio (time expansion ratio) is compared with R _V, and the compression (expansion) of the voice section is emphasized as much as possible, and the compression (expansion) of the voice section has a large time compression rate (time expansion rate). It's a method that is only done sometimes.

[0009]

By this method, the compression ratio (expansion ratio) for the silent section and the voice section in the block is determined for each R _V. When R _V is 1.0, it is a 100% voice section, and when it is 0.0, it is a 100% voice section. First, in the case of time compression, when a block contains many silent intervals and k · R _V is smaller than the time compression ratio R _C , the desired time compression ratio can be achieved simply by compressing the silent intervals. When there are few silent sections in the block and k · R _V is larger than R _C , the desired time compression ratio cannot be achieved only by compressing the silent sections, so the speech section is also compressed at the same time. Where the constant k
Is a value greater than 1 and determines the margin for silence interval compression.

Next, in the case of time extension, there are not many silent sections in the block and k / R is calculated from the time extension ratio R _{E.
When V} is small, the desired time extension ratio can be achieved simply by extending the silent section, and there are many silent sections in the block.
When k / R _V is larger than R _E, there is a limit in achieving the desired time extension ratio by only the extension of the silent section, so the extension of the voice section is also performed at the same time. Here, the constant k is a value smaller than 1 and determines the margin of the silent section expansion.

By performing the control as described above, it is possible to hear the utterance as natural as possible even if the reproduction speed is changed, and it is possible to obtain the reproduced sound in which the processing for the voice section is small and the processing distortion is reduced.

[0012]

First Embodiment A first embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 shows a flowchart in the first embodiment of the present invention. The symbols used in (FIG. 1) will be described. R _V is the ratio of the sum of the voice section lengths in a block to the entire block length, as shown in (Equation 2) below.

[0014]

[Equation 2]

R _C is a compression ratio for time-compressing the input signal as shown in the following (Equation 3).

[0016]

[Equation 3]

The total time length of the block length is T _B , the total time length of the silent section is T _S , and the total time length of the voice section is
T _V, the compression ratio of time compression on silence section C _S, the compression ratio of time compression on the voice section and C _V. These, T _B , T _S ,
Writing R _C by T _V , C _S , C _V gives the following (Equation 4).

[0018]

[Equation 4]

The relational expression between T _B and T _S , T _V is as shown in the following (Equation 5).

[0020]

[Equation 5]

Next, the flowchart of FIG. 1 will be described. First, the input signal is read into the buffer memory for one block. Then, with respect to the signal in this block, the voice section and the silent section are determined to obtain R _V. Then, a comparison of a constant k (k> 1) k · multiplied by the R _V and R _C to the R _V, if the R _C ≧ k · R _V and, in the case of R _C <k · R _V of
Calculate the values of C _S and C _V as in (Table 1) below.

[0022]

[Table 1]

Here, when the value of R _V is small, that is, when the ratio of the silent section in the block is large, R _C ≧ k · R _V
Since the condition is easily satisfied, the time compression of the voice section is not performed, and the target time compression ratio is realized only by the time compression of the silent section. On the contrary, when there are not many silent sections in the block, the condition of R _C <k · R _V is easily satisfied, and the silent section or the voice section is time-compressed. In addition, R _V
There are two cases depending on the value of. Those in which the change in the value of C _S and C _V for R _V in the graph (FIGS. 2 (a)
(B)). As is clear from this figure, when R _V changes from 0 to 1, the compression state for the silent section or the speech section is as follows.First, only the silent section is compressed, and then the silent section and the speech section. Both of the compressions are performed, and in the vicinity of 1, only the voice section is compressed.

As shown in (Table 1), after determining C _S and C _V , the data in the block is processed. The silent section is compressed based on the compression ratio C _S , and the voice section is compressed based on the compression ratio C _V. With the above, the processing for one block is completed, and it is confirmed whether the same processing is performed for the next block.

As described above, according to this embodiment, the compression ratios for the silent section and the voice section are changed according to the value of R _{V in} the current block. Then, in a block with many silent sections, the speech section is not compressed, but only the silent section is compressed to the desired compression ratio R _C , so that there is no processing distortion in the speech section and natural high-speed reproduction is performed. Can be realized.

A second embodiment of the present invention will be described below with reference to the drawings. FIG. 3 shows a flowchart in the second embodiment of the present invention. The symbols used in (FIG. 3) will be described. R _E is an expansion ratio for time-expanding the input signal as shown in (Equation 6) below.

[0027]

[Equation 6]

Further, R _V , T _B , T _S and T _V are the same as those in the first embodiment. _Let E _{S be} the expansion ratio of the time expansion for the silent interval and E _{V be} the expansion ratio of the time expansion for the voice interval. these,
When R _E is written with T _B , T _S , T _V , E _S , and E _V , the following (Equation 7) is obtained.

[0029]

[Equation 7]

Next, the flowchart of FIG. 3 will be described. First, the input signal is read into the buffer memory for one block. Then, with respect to the signal in this block, the voice section and the silent section are determined to obtain R _V. Then, k / R _V obtained by multiplying the reciprocal of R _{V by} a constant k (k <1)
By comparing with R _E , the values of E _S and E _V when R _E ≧ k · R _V and when R _E <k · R _V are calculated as shown in (Table 2) below.

[0031]

[Table 2]

Here, when the value of R _V is large, that is, when the ratio of the silent section in the block is small, R _E ≧ k / R
The condition of _V is easily satisfied, the time extension of the voice section is not performed, and the target time extension ratio is realized only by the time extension of the silent section. On the contrary, when there are many silent sections in the block, the condition of R _E <k / R _V is easily satisfied, and the silent section or the voice section is time-expanded in parallel.
As shown in (Table 2), after determining E _S and E _V , the data in the block is processed. The silent section is expanded based on the expansion ratio E _S , and the voice section is expanded based on the expansion ratio E _V. With the above, the processing for one block is completed, and it is confirmed whether the same processing is performed for the next block.

As described above, according to this embodiment, the expansion ratio for the silent section and the voice section is changed according to the value of R _{V in} the current block. In a block with few silent sections, the sound section is not expanded, but only the silent section is expanded to the desired expansion ratio R _E , so there is no processing distortion with respect to the sound section and a natural low-speed playback is performed. Can be realized.

[0034]

As described above in detail, according to the present invention, the ratio R _V of the voice section length to the entire block length in a predetermined block length is
And compare this value with the desired compression ratio (expansion ratio),
Depending on the result, (1) the entire time compression (time expansion) is performed by compressing (expanding) only the silent section, or (2) the independent compression ratio (expansion ratio) for the silent section and the voice section. This is a method of performing audio reproduction by selecting whether to install and perform time compression (time expansion). With this method, for high-speed playback, for blocks containing many silent intervals, time compression is achieved only by compressing silent intervals, and for low-speed playback, for blocks with few silent intervals. Is to extend time only by extending the silent section. Therefore, when no processing is applied to the voice section of the original signal, processing distortion is reduced. Another object of the present invention is to provide a voice reproduction method that allows humans to naturally hear voices by deleting unnecessary silence intervals during high-speed reproduction and inserting silence intervals that allow time to think during low-speed reproduction.

[Brief description of drawings]

FIG. 1 is a flowchart of an audio reproducing method according to a first embodiment of the present invention.

FIG. 2A shows R _V and C _S in the first embodiment of the present invention.
It is a graph which shows the relationship of. (B) is a graph showing the relationship between R _V and C _v in the first embodiment of the present invention.

FIG. 3 is a flowchart of an audio reproducing method according to a second embodiment of the present invention.

FIG. 4 is a flowchart of a conventional audio reproduction method.

Claims (2)

[Claims] 1. A voice section or a silent section is determined within a predetermined block length range, a ratio R _V between the voice section length and the entire block length is obtained, and k · R is calculated from a target time compression ratio R _{C. V}
When (k is a constant of 1 or more) is small or equal,
Compressing the time length by compressing only the silent section, from R _C
When k · R _V takes a large value, a voice reproduction method for performing high-speed voice reproduction by performing compression at a predetermined compression ratio for a silent section or a voice section to compress a time length and repeating the above processing. 2. A voice section or a silent section is determined within a predetermined block length range, a ratio R _V between the voice section length and the entire block length is obtained, and k / R is calculated from a target time expansion ratio R _E.
When _V (k is a constant of 1 or less) is smaller or equal value, extends the length of time by extending the silent section, from R _E
When k / R _V takes a large value, a sound reproduction method in which a sound section or a sound section is expanded at a predetermined expansion ratio to expand the time length, and the above processing is repeated to reproduce the sound at a low speed.