JP2532731B2 - Voice speed conversion device and voice speed conversion method - Google Patents

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice speed conversion device and a voice speed conversion method for changing only a duration without changing the fundamental frequency of voice.

2. Description of the Related Art Conventionally, an audio speed conversion device has been used to perform early or late listening of an audio signal recorded on a tape recorder or the like.

Hereinafter, the conventional audio speed converter as described above will be described with reference to the drawings.

FIG. 9 shows the configuration of a conventional voice speed conversion device. In FIG. 9, 91 is an A / D converter, 92 is a buffer, 93 is a speed control circuit, 94 is a data reading circuit, 95 is a muting circuit, and 96 is a D / A converter.

Regarding the audio speed conversion device configured as described above,
The operation will be described below.

First, the input signal is converted into a digital signal by the A / D converter 91 and written in the buffer 92. Next, the speed control circuit 93 controls the data reading circuit 94 according to the speed conversion ratio to read the data from the buffer 92. The reproduction speed can be variously changed by the data reading method.
When increasing the reproduction speed, the data to be read is thinned out in block units. When the reproduction speed is slowed down, the data to be read is repeated in block units. The muting circuit 95 mutes the discontinuous portion between the blocks, and the D / A converter 96 converts the muted signal into an analog signal for output.

FIG. 10 schematically shows the case where the speed conversion ratio α is 0.5 and 2.0. Here, the speed conversion ratio α is defined.

In the figure, (a) is an original sound, (b) is an audio signal converted at a speed conversion ratio = 0.5, and (c) is an audio signal converted at a speed conversion ratio = 2.0.

Problems to be Solved by the Invention However, in the configuration as described above, in the case of increasing the speed, intelligibility of consonants and the like deteriorates in order to thin out data, and since discontinuous points of the waveform are muted. In addition, there is a problem that only a voice with poor naturalness can be obtained.

Also, the method using the pitch period is not suitable because it is difficult to extract the pitch when music or noise is superimposed, and is not appropriate.

In view of the above problems, the present invention has less discontinuity in waveform,
(EN) Provided are a voice speed conversion device and a voice speed conversion method capable of outputting a highly natural voice that does not cause data loss.

Means for Solving the Problems In order to achieve this object, a voice speed conversion device of the present invention comprises a first memory for storing voice, a second memory for storing voice, the contents of the first memory, and the content of the first memory. A correlator that calculates a correlation function of the contents of the second memory; a window function generator that outputs a window function; and a first multiplier that multiplies the contents of the first memory by the output of the window function generator. A second multiplier that multiplies the contents of the second memory by the output of the window function generator, and the output of the first multiplier and the output of the second multiplier based on the output of the correlator. Is composed of an adder that performs addition at a position where the value of the correlation function is large, and a selection circuit that switches between the original voice and the output of the adder.

Also, in the voice speed conversion method of the present invention, the speed conversion ratio is α, the correlation function of the first waveform of the time length T and the second waveform of the time length T is calculated, and the correlation function is calculated for the second waveform. The first waveform is shifted to a position where the value of the correlation function is large, and the first waveform is multiplied by a window function whose amplitude gradually increases, and the second waveform is multiplied by a window function whose amplitude gradually decreases. 1 waveform and the second waveform are added and output, and a third waveform having a time length {T / (α-1) + Tc) following the first waveform is output, and in the next processing The starting point of the first waveform is the first
The starting point of the waveform of is the point delayed by {T / (α-1)}, and the starting point of the second waveform in the next process is the first point.
The start point of the waveform of is set to a point delayed by {αT / (α-1) + Tc}, and the voice reproduction time is changed to 1.0 times or more of the original sound by repeating all the above processes. It is a thing.

Further, in the voice speed conversion method of the present invention, the speed conversion ratio is set to α
Then, the correlation function of the first waveform having the time length T and the second waveform having the time length T is calculated, and the second waveform is displaced from the first waveform by Tc to a position having a large value of the correlation function. , The first
Is multiplied by a window function whose amplitude is gradually reduced, the second waveform is multiplied by a window function whose amplitude is gradually increased, and the first waveform and the second waveform are added and output. A third waveform having a time length {(2α-1) T / (1-α) -Tc} following the second waveform is output, and the starting point of the first waveform in the next process is the second waveform. The starting point is a point delayed by {αT / (1-α) -Tc}, and the starting point of the second waveform in the next processing is the starting point of the second waveform {T / (1-α) }, And the voice reproduction time is changed in the range of 0.5 times to 1.0 times the original sound by repeating all the above processes.

Further, in the voice speed conversion method of the present invention, the speed conversion ratio is set to α
Then, the start point of the second waveform is initialized to a point obtained by delaying the start point of the first waveform by {(1-α) T / α}, and the first waveform of the time length T and the time length T Of the second waveform of the second waveform is calculated, and the second waveform is shifted from the first waveform to a position where the value of the correlation function is large by Tc, and the first waveform has a window function whose amplitude gradually decreases. And the second waveform is multiplied by a window function whose amplitude increases gradually to obtain the first waveform and the second waveform.
When the Tc is negative, the added waveform is output, and then the third waveform of the time length −Tc following the second waveform is output, while when Tc is not negative, Only the section of the time length T from the beginning of the added waveforms is output, and the starting point of the first waveform in the next processing is the point obtained by delaying the starting point of the second waveform by {T-Tc}. The starting point of the second waveform in the processing of is the same as the starting point of the second waveform {T /
It is characterized in that the audio reproduction time is changed to 0.5 times or less of the original sound by repeating all the processes other than the above-mentioned initial setting.

Effect According to the present invention, with this configuration, the window function is multiplied, and the voice waveform is superposed and added so that the value of the correlation function becomes large, so that the discontinuity of the amplitude and phase of the waveform and the loss of data are not caused much. You can convert the speed of voice.

Embodiment An embodiment of the present invention will be described below with reference to the drawings.

The present invention has few amplitude and phase discontinuities in the waveform,
It is an object of the present invention to provide a voice speed conversion device capable of outputting a voice with rich naturalness, which does not cause data loss so much, and which can be realized by simple hardware.

FIG. 1 shows a configuration diagram of an audio speed conversion apparatus according to an embodiment of the present invention. In FIG. 1, 11 is A / D
Converter, 12 buffer, 13 speed control circuit, 14 demultiplexer, 15 first memory, 16 second memory, 17
Is a correlator, 18 is a window function generator, 19 is a first multiplier, 20 is a second multiplier, 21 is an adder, 22 is a multiplexer, and 23 is
It is a D / A converter.

Regarding the audio speed conversion device configured as described above,
The operation will be described below.

First, the input signal is converted into a digital signal by the A / D converter 11 and written in the buffer 12. Next, speed control circuit
13 controls the demultiplexer 14 according to the speed conversion ratio,
The data in the buffer 12 is stored in the first memory 15 and the second memory.
16 or output to the multiplexer 22. And the first
The correlator 17 calculates a correlation function between the contents of the memory 15 and the contents of the second memory 16, and outputs the information to the speed control circuit 13, the window function generator 18 and the adder 21. Then, the window function generator 18 generates a window function that gradually increases or decreases based on the information from the correlator 17 and the speed conversion ratio, outputs the window function to the first multiplier 19, and outputs a window function complementary to the window function. Second multiplier 20
Output to. The first multiplier 19 then multiplies the contents of the first memory 15 by the output of the window function generator 18, the second multiplier 20 the contents of the second memory 16 and the window function generator 18. Multiply with the output of. Based on the information from the correlator 17, the adder 21 shifts the output of the first multiplier 19 and the output of the second multiplier 2 to the position where the correlation function has a large value, and performs the addition. To output. Finally, the multiplexer 22 selects the output of the adder 21 and the output of the demultiplexer 14 based on the control signal from the speed control circuit 13 to select D /
It is output to the A converter 23 and converted into an analog signal.

FIG. 2 schematically shows the case where the speed conversion ratio α is 0.5 and 2.0.

In FIG. 2, (a) is an original sound, (b) is a voice signal converted at a speed conversion ratio = 0.5, and (c) is a speed conversion ratio = 2.0.
Is the audio signal converted by
n indicates the front part and the rear part of the voice, respectively.

As described above, according to the present embodiment, the contents of the first memory 15 and the contents of the second memory 16 are multiplied by the complementary window function output from the window function generator 18 that gradually decreases or increases. The multiplier 19 and the second multiplier 20 are used for multiplication, and the adders 21 add the outputs of the respective multipliers, so that there is little discontinuity in the amplitude of the waveform and there is little natural loss of data. It can output rich sound. Then, the correlator 17 calculates the correlation function between the contents of the first memory 15 and the contents of the second memory 16, and the adder 21 outputs the output of the first multiplier 19 and the output of the second multiplier 20. By shifting to a position where the value of the correlation function becomes large and performing the addition, it is possible to output high-quality speech with little phase discontinuity of the waveform. Further, the section length for outputting the signal as it is without superimposing and adding the signal is set by the speed control circuit 13 and the demultiplexer 14.
And the multiplexer 22 adjust the speed conversion ratio easily, and at the same time, it is possible to quickly absorb the shift in the speed conversion ratio caused by performing the addition by shifting to a position where the value of the correlation function becomes large. It will be possible.

A first embodiment of the voice speed conversion method of the present invention will be described below with reference to the drawings.

The present invention has few amplitude and phase discontinuities in the waveform,
It is intended to provide a voice speed conversion method capable of outputting a highly natural voice that does not cause data loss in a range of a speed conversion ratio α ≧ 1.0.

FIG. 3 shows a flowchart of the voice speed conversion method in this embodiment.

First, the A pointer is set to 0 and the B pointer is set to T. Next, the waveform (XA) of the T section is input from the A pointer. Then, the waveform (XB) of the T section is input from the B pointer. Next, the A pointer plus T is set as the B pointer. Then, the correlation function of XA and XB is calculated. Next, based on the previously obtained correlation function, XA is provided with a gradually increasing window, and based on the previously obtained correlation function, XB is provided with a gradually decreasing window. Next, based on the correlation function obtained earlier, XA
And XB are shifted by Tc to the position where the value of the correlation function is large, and then added. When T-Tc is less than αT / (α-1), all the added waveforms are output, and the waveform (XC) in the T / (α-1) + Tc section is further output from the B pointer. On the other hand, when αT / (α-1) is less than T-Tc, the added waveform is output only in the αT / (α-1) section. Next, add T / (α-1) + Tc to the B pointer, and add T / (α- to the A pointer.
Add 1). Next, return to the third step from the beginning. The speed conversion ratio α is α ≧ 1.

FIG. 4 schematically shows a case where the voice speed is converted by the speed conversion ratio α of 2.0 and 3.0 based on the above conversion method.

In FIG. 4, (a) is an original sound, (b) is a voice signal converted at a speed conversion ratio = 2.0, and (c) is a speed conversion ratio = 3.0.
The audio signal converted by is shown.

As described above, according to the present embodiment, by multiplying XA and XB by a window function that gradually increases or decreases complementarily, outputs a waveform in which they are added, and by repeating the operation of outputting XC subsequent to XA, It is possible to output a highly natural voice in which the discontinuity of the waveform amplitude is small and the data is not lost so much in the range of the speed conversion ratio α ≧ 1.0. And XA
By calculating the correlation function between XB and XB and shifting the value to a position where the value of the correlation function becomes large and performing addition, it is possible to output high-quality speech with little discontinuity in the phase of the waveform.
Furthermore, the signals are output as they are without being superimposed and added.
The speed conversion ratio can be easily changed by adjusting the section length of XC, and at the same time, the speed conversion ratio deviation due to the addition performed by shifting to the position where the value of the correlation function becomes large can be absorbed promptly. Is possible.

A second embodiment of the voice speed conversion method of the present invention will be described below with reference to the drawings.

The present invention has few amplitude and phase discontinuities in the waveform,
It is intended to provide a voice speed conversion method capable of outputting a highly natural voice that does not cause data loss in a range of a speed conversion ratio of 0.5 ≦ α ≦ 1.0.

FIG. 5 shows a flowchart of the voice speed conversion method in this embodiment.

First, the A pointer is set to 0 and the B pointer is set to T. Next, the waveform of the T section (XA) is input from the A pointer, and the waveform of the T section (XB) is input from the B pointer. Next, the B pointer plus T is set as the A pointer, and the correlation function of XA and XB is calculated. Next, based on the previously obtained correlation function, XA is provided with a gradually decreasing window, and further, based on the previously obtained correlation function, XB is provided with a gradually increasing window. Next, based on the previously obtained correlation function, XA and XB are shifted by Tc to a position where the value of the correlation function is large, and then added. Then, when T + Tc is less than αT / (1-α), all the added waveforms are output, and from the A pointer, (2α-
1) Output the waveform (XC) in the T / (1-α) -Tc section. On the other hand, when αT / (1-α) is less than T + Tc, the added waveform is output only in the αT / (1-α) section. Next, (2α-1) T / (1-α) -Tc is added to the A pointer, and T / (1-α) is added to the B pointer. Next, return to the third step from the beginning.

FIG. 6 schematically shows the case where the voice speed is converted at the speed conversion ratio α of 2/3 and 0.5 based on the above conversion method.

In FIG. 6, (a) is an original sound, (b) is a voice signal converted at a speed conversion ratio = 2/3, and (c) is a speed conversion ratio = 0.5.
The audio signal converted by is shown.

As described above, according to the present embodiment, XA and XB are complementarily multiplied by a gradually decreasing or gradually increasing window function, and a waveform obtained by adding them is output, by repeating the operation of outputting XC subsequent to XB, It is possible to output a highly natural voice in which the discontinuity of the waveform amplitude is small and the data is not often lost in the range where the speed conversion ratio is 0.5 ≦ α ≦ 1.0. Then, the correlation function between XA and XB is calculated, and the addition is performed by shifting to a position where the value of the correlation function becomes large, whereby high-quality speech can be output with little discontinuity in the phase of the waveform. Furthermore, by adjusting the section length of XC that is output as it is without superimposing and adding signals, it is possible to easily change the speed conversion ratio, and at the same time, perform the addition by shifting to a position where the value of the correlation function becomes large. It is possible to quickly absorb the shift in the speed conversion ratio.

Hereinafter, the third embodiment of the speed conversion method of the present invention will be described.
This will be described with reference to the drawings.

The present invention has few amplitude and phase discontinuities in the waveform,
The present invention provides a voice speed conversion direction capable of outputting a sound with rich naturalness in a speed conversion ratio range of α ≦ 0.5.

FIG. 7 shows a flowchart of the voice speed conversion method in this embodiment.

First, set the A pointer to 0 and the B pointer to (1-α) T /
Set to α. Next, the waveform (X
A) is input, and the waveform (XB) of the T section is input from the B pointer. Next, the B pointer plus T is set as the A pointer, and the correlation function of XA and XB is calculated. Next, based on the previously obtained correlation function, XA is provided with a gradually decreasing window, and further, based on the previously obtained correlation function, XB is provided with a gradually increasing window. Next, based on the previously obtained correlation function, XA and XB are shifted by Tc to a position where the value of the correlation function is large, and then added. When Tc is negative, all the added waveforms are output, and the waveform (XC) in the -Tc section is output from the A pointer. On the other hand, when Tc is not negative, the added waveform is output only in the T section. Next, -Tc is added to the A pointer, and T / α is added to the B pointer. Next, return to the third step from the beginning.

FIG. 8 schematically shows a case where the voice speed is converted at the speed conversion ratio α of 1/3 and 1/4 based on the above conversion method.

In FIG. 8, (a) is an original sound, (b) is a voice signal converted at a speed conversion ratio = 1/3, and (c) is a speed conversion ratio = 1/4.
The audio signal converted by is shown.

As described above, according to the present embodiment, XA and XB are complementarily multiplied by a gradually decreasing or gradually increasing window function, and a waveform obtained by adding them is output, by repeating the operation of outputting XC subsequent to XB, It is possible to output a highly natural voice with less discontinuity in the amplitude of the waveform in the range where the speed conversion ratio is α ≦ 0.5. Then, the correlation function between XA and XB is calculated, and the addition is performed by shifting to a position where the value of the correlation function becomes large, whereby high-quality speech can be output with little discontinuity in the phase of the waveform. Further, by adjusting the position of the B pointer with respect to the A pointer, the speed conversion ratio can be easily changed, and at the same time, the shift of the speed conversion ratio due to the addition by shifting to the position where the value of the correlation function becomes large can be promptly performed. It becomes possible to absorb it.

EFFECTS OF THE INVENTION As described above, in the voice speed conversion device of the present invention, the complementary two gradually decreasing or gradually increasing window functions output from the window function generator are added to the contents of the first memory and the contents of the second memory. By performing multiplication using the first multiplier and the second multiplier and adding the outputs of the respective multipliers by the adder, there is little discontinuity in the amplitude of the waveform, and there is little loss of data. A rich voice can be output. Further, the correlator calculates the correlation between the contents of the first memory and the contents of the second memory, and the adder compares the output of the first multiplier and the output of the second multiplier with the value of the correlation function. By performing the addition while shifting to a position where the waveform becomes large, it is possible to output a high-quality voice with little discontinuity in the phase of the waveform. Further, the speed control circuit, demultiplexer, and multiplexer adjust the section length to output the signal as it is without superimposing and adding the signal, so that the speed conversion ratio can be easily changed and at the same time the value of the correlation function becomes large. Not only can it be possible to quickly absorb the shift in the speed conversion ratio due to shifting and addition, but also when the speed conversion ratio α that is relatively often used is near 1.0, the section length to be output as it is becomes longer, It is possible to obtain the effect that the sound quality is less deteriorated.

Further, the voice speed conversion method of the present invention multiplies the first waveform and the second waveform by the number of correlations which gradually increase or decrease complementarily, and outputs the waveform obtained by adding them to obtain the first or second waveform. By repeating the operation of outputting the subsequent third waveform, it is possible to output a voice having a high degree of naturalness, with little discontinuity in the amplitude of the waveform, and causing little data loss.

Furthermore, the correlation function between the first waveform and the second waveform is calculated, and the addition is performed by shifting the correlation function to a position where the value of the correlation function becomes large. Can be output.

[Brief description of drawings]

FIG. 1 is a block diagram of a voice speed conversion apparatus according to an embodiment of the present invention, FIG. 2 is a schematic diagram of processing waveforms by the voice speed conversion apparatus according to an embodiment of the present invention, and FIG. 3 is a first view of the present invention.
A flowchart of a voice speed conversion method in the embodiment of
FIG. 4 is a schematic diagram of a processing waveform by the voice speed converting method in the first embodiment of the present invention, FIG. 5 is a flowchart of the voice speed converting method in the second embodiment of the present invention, and FIG.
FIG. 7 is a schematic diagram of a waveform processed by the voice speed converting method according to the second embodiment of the present invention, FIG. 7 is a flow chart of the voice speed converting method according to the third embodiment of the present invention, and FIG. FIG. 9 is a schematic diagram of a processed waveform by a voice speed conversion method in the third embodiment, FIG. 9 is a configuration diagram of a conventional voice speed conversion device, and FIG. 10 is a schematic diagram of a processed waveform by a conventional voice speed conversion device. 13 …… Speed control circuit, 14 …… Demultiplexer, 15 ……
First memory, 16 ... Second memory, 17 ... Correlator, 18
...... Window function generator, 19 ...... First multiplier, 20 …… Second multiplier, 21 …… Adder, 22 …… Multiplexer.

Claims (6)

(57) [Claims] 1. A first memory for storing voice, a second memory for storing voice, contents of the first memory and the second memory.
A correlator that calculates a correlation function with the contents of the memory, a window function generator that outputs a window function, and a first multiplier that multiplies the contents of the first memory by the output of the window function generator, A second multiplier for multiplying the contents of the second memory by the output of the window function generator; and the first multiplier based on the output of the correlator.
An adder for adding the output of the multiplier and the output of the second multiplier at a position where the value of the correlation function is large, and a selection circuit for switching between the original voice and the output of the adder. A voice speed conversion device. 2. A velocity conversion ratio is α, and a correlation function of a first waveform having a time length T and a second waveform having a time length T is calculated, and the first waveform is calculated with respect to the second waveform. The position where the value of the correlation number is large is shifted to Tc, the first waveform is multiplied by a window function whose amplitude is gradually increased, and the second waveform is multiplied by a window function whose amplitude is gradually decreased. The second waveform is added and output, and the time length {T / (α-1) + that follows the first waveform is added.
The third waveform of Tc} is output, and the start point of the first waveform in the next processing is the start point of the first waveform {T / (α-
1)} and the second point in the next processing
The starting point of the waveform of is the same as the starting point of the first waveform {αT / (α
−1) + Tc}, and by repeating all the above processes, the voice reproduction time is 1.0
Voice speed conversion method that changes more than twice. 3. The time length of the added waveform is {αT / (α-1)} when the first waveform and the second waveform are added and output.
If it exceeds, the time length from the beginning of the added waveform is {α
The voice speed conversion method according to claim 2, wherein only the T / (α-1)} section is output and the third waveform is not output. 4. A velocity conversion ratio is α, and a correlation function between a first waveform having a time length T and a second waveform having a time length T is calculated, and the second waveform is calculated with respect to the first waveform. The position where the value of the correlation function is large is shifted to Tc, the first waveform is multiplied by a window function whose amplitude is gradually reduced, and the second waveform is multiplied by a window function whose amplitude is gradually increased. The second waveform is added and output, and the time length following the second waveform {(2α-1)
The third waveform of T / (1-α) -Tc} is output, and the starting point of the first waveform in the next processing is {αT / (1-α) -Tc. }, And the starting point of the second waveform in the next processing is the point of delaying the starting point of the second waveform by {T / (1-α)}. A voice speed conversion method that changes the playback time of voice in the range of 0.5 to 1.0 times the original sound by repeating. 5. The time length of the added waveform is {αT / (1-α)} when the first waveform and the second waveform are added and output.
If it exceeds, the time length from the beginning of the added waveform is {α
The voice speed conversion method according to claim (4), wherein only the T / (1-α)} section is output and the third waveform is not output. 6. The velocity conversion ratio is α, the starting point of the second waveform is initialized to a point obtained by delaying the starting point of the first waveform by {(1-α) T / α}, and the time length is set. First waveform of T and time length T
Of the second waveform of the second waveform is calculated, and the second waveform is shifted from the first waveform to a position where the value of the correlation function is large by Tc, and the first waveform has a window function whose amplitude gradually decreases. And multiplying the second waveform by a window function of increasing amplitude,
The first waveform and the second waveform are added, and when Tc is negative, the added waveform is output, and then the third waveform having a time length −Tc following the second waveform is output. On the other hand, when Tc is not negative, only the section of the time length T from the beginning of the added waveform is output, and the start point of the first waveform in the next processing is the start point of the second waveform {T-Tc. }, And the start point of the second waveform in the next processing is the second point.
The voice speed conversion method in which the start point of the waveform is delayed by {T / α}, and the reproduction time of the voice is changed to 0.5 times or less of the original sound by repeating all processes other than the initial setting.