JPH06222794A - Voice speed conversion method - Google Patents

Abstract

PURPOSE:To provide a voice conversion method capable of outputting a voice with the little discontinuity of a waveform and with the little omission of data and rich in a natural property by adding them after multiplying a first signal and a second signal by a window function. CONSTITUTION:The first signal with a time length T is inputted from an input pointer. The second signal with the time length T is inputted succeeding to that. The correlative function between the first signal and the second signal is calculated, and a time delay Tc when the value of the correlative function is maximum is retrieved. The first signal is multiplied by the window function gradually increasing an amplitude according to the obtained time delay Tc. The second signal is multiplied by the window function gradually decreasing the amplitude according the time delay Tc. After the first signal multiplied by the window function and the second signal multiplied by the window function are shifted to the position of the time delay Tc where the correlative function becomes a maximum, they are added. Then, the obtained signal and a third signal succeeding to the first signal are outputted only for a interval alpha(T-Tc)/(alpha-1) (where alpha is a time base transformation ratio).

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice speed conversion method in which only the duration is changed without changing the fundamental frequency of voice.

[0002]

2. Description of the Related Art In recent years, a voice speed conversion device has been used to perform fast listening or slow listening of a voice signal recorded on a tape recorder or the like.

A conventional voice speed conversion device as described above will be described below with reference to the drawings.

FIG. 8 shows the structure of a conventional voice speed converter. In FIG. 8, 81 is an A / D converter, and 8
2 is a buffer, 83 is a speed control circuit, 84 is a data reading circuit, 85 is a muting circuit, and 86 is a D / A converter.

The operation of the voice speed conversion device configured as described above will be described below.

First, the analog input signal is the A / D converter 8
At 1, it is converted to a digital signal and written in the buffer 82. Next, the speed control circuit 83 controls the data reading circuit 84 according to the time base conversion ratio to read the data from the buffer 82. By such a reading method, the reproduction speed can be variously changed. When shortening the reproduction time, the data to be read is thinned out in block units. When lengthening the reproduction time, the data to be read is repeated in block units. The muting circuit 85 mutes the discontinuous portion between the blocks, and the D / A converter 86 converts the muted signal into an analog signal for output.

FIG. 9 is a diagram showing a time-series relationship of signals between an original sound and a sound obtained by time-base conversion of the original sound.
(B) is a signal converted with the time-axis conversion ratio α = 0.5,
(C) schematically shows a signal converted at the time axis conversion ratio α = 2.0. Here, the time axis conversion ratio α is defined by the following equation.

[0008]

[Equation 1]

[0009]

However, in the above configuration, when the time axis is compressed to increase the speed, consonants and the like are lost to thin out the data, resulting in a decrease in clarity and further block. The connection point of is discontinuous, and although the connection point is muted in order to reduce it, there is a problem that only amplitude and phase discontinuous voices with poor naturalness can be obtained.

In another conventional voice speed conversion device,
There is also a method of using the pitch period of the input signal such as TDHS (Time Domein Harmonic Scaling), but when music or noise is superimposed on the input signal, it is difficult to extract the pitch, so it is not applicable. There wasn't.

In view of the above problems, the present invention provides a voice speed conversion method capable of outputting a natural voice having few waveform discontinuities and causing little data loss.

[0012]

In order to achieve this object, a voice speed conversion method of the present invention comprises a first signal having a finite time length T and a second signal having a finite time length T following the first signal. A time delay Tc at which the value of the correlation function is maximized is calculated by calculating a correlation function with the signal, and based on the time delay Tc at which the value of the correlation function is maximized for the first signal and the second signal. And a second signal obtained by multiplying the window function and the second signal obtained by multiplying the window function, the amplitudes of which are complementary to each other. At the position of the time delay Tc, a third signal is continuously output to the added signal, and the added signal and the third signal are converted to the time axis conversion ratio α (= output time / input time ) And the time delay Tc at which the value of the correlation function becomes maximum and the finite time length T Output only during long delays a first signal and a time value of the correlation function the starting point and α time axis conversion ratio of the second signal becomes maximum in the next processing Tc and finite time length T
The voice speed conversion method is characterized in that the reproduction time of the voice is changed with respect to the length of the original sound by repeating all the processes described above.

[0013]

According to such a method, the first signal and the second signal are multiplied by the window function and then added to reduce the missing of the added signal and the discontinuity of the amplitude, and also to multiply the window function. By adding the first signal and the second signal obtained by multiplying the window function at the position of the time delay Tc where the value of the correlation function is maximum, the discontinuity of the phase is reduced.

Furthermore, the signal obtained by adding the first signal multiplied by the window function and the second signal multiplied by the window function and the third signal following the added signal are correlated with the time axis conversion ratio α. By outputting only the time length determined based on the time delay Tc that maximizes the value of the function and the finite time length T, it is possible to reduce the loss of the signal and perform conversion to an arbitrary speed.

[0015]

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

According to the present invention, it is possible to output a highly natural voice in which there is little discontinuity in the amplitude and phase of the signal and there is no loss of data in the range where the time axis conversion ratio α is α ≧ 1.0. The present invention provides a possible voice speed conversion method.

Here, the time axis conversion ratio α is defined by the following equation.

[0018]

[Equation 2]

FIG. 1 shows a flow chart of a voice speed converting method in the first embodiment of the present invention. The operation will be described below.

First, in step 11, the input pointer is reset. Next, in step 12, the first signal (X _A ) in the T section is input from the input pointer. Then, in step 13, T is added to the input pointer. Next, at step 14, the second signal (X _B ) in the T section is input from the input pointer.

In step 15, the correlation function of the first signal X _A and the second signal X _B is calculated, and the time delay Tc at which the value of this correlation function becomes maximum is searched for. Then in step 16,
The first signal X _A is multiplied by a window function whose amplitude gradually increases, based on the time delay Tc at which the correlation function obtained above becomes maximum. Then, in step 17, the second signal X _B is multiplied by the window function whose amplitude is gradually reduced, based on the time delay Tc at which the correlation function obtained previously becomes maximum. Next, at step 18, the first signal multiplied by the window function and the second signal multiplied by the window function are shifted to the position of the time delay Tc at which the correlation function becomes maximum and then added. Then, in step 19, the signal added in step 18 and the signal following the first signal X _A , that is, the third signal (X
_C ) is output only in the α (T-Tc) / (α-1) section. Next, at step 20, the input pointer is set to (2T-αT
-Tc) / (α-1) is added. After this, step 12
Return to.

In FIG. 2, the correlation function of the first signal X _A and the second signal X _B in step 15 shown in FIG. 1 is calculated, and the time delay Tc at which the value of the correlation function becomes maximum is searched for. It shows a flowchart of processing. The operation will be described below.

First, steps 201, 202 and 20
At 3, the time delay τ, the time delay Tc that maximizes the value of the correlation function, and the maximum value Rmax of the correlation function are initialized to zero.
Next, in step 204, when the time delay τ is not negative as shown in (Equation 3), the first signal X _A and the second signal X _B
The correlation function R (τ) of is calculated.

[0024]

[Equation 3]

Then, in step 205, if the correlation function R (τ) obtained in step 204 is not larger than the maximum value Rmax of the correlation function obtained before that, the process branches to step 208. In step 206, the maximum value Rmax of the correlation function is updated to R (τ), and in step 207, the time delay Tc at which the value of the correlation function is maximum is updated to τ. Next, at step 208, the time delay τ is increased by one point. Then, in step 209, if the time delay τ does not exceed τ max ₊ , the process returns to step 204, and the processing from steps 204 to 208 is repeated until the time delay τ exceeds τ max ₊ . When the above condition is satisfied, the time delay τ is initialized to -1 in step 210. Next, at step 211, the correlation function R (τ) of the first signal X _A and the second signal X _B when the time delay τ is negative as shown in (Equation 4) is calculated.

[0026]

[Equation 4]

Then, in step 212, step 211
If the correlation function R (τ) obtained in step 1 is not larger than the maximum value Rmax of the correlation function obtained before that, the process branches to step 215, and if not, step 21
In step 3, the maximum value Rmax of the correlation function is updated to R (τ), and in step 214, the time delay Tc at which the value of the correlation function becomes maximum is updated to τ. Next, at step 215, the time delay τ is decreased by one point. Then, in step 216, the time delay τ
There .tau.max _- returns to step 211 if no smaller than,
In the processing from steps 211 to 215, the time delay τ is τ
max _- repeated until than smaller. Finally step 21
At 7, the time delay Tc that maximizes the value of the correlation function is output.

FIG. 3 shows steps 16 and 17 shown in FIG.
It is a schematic diagram of the process in 18 and.

FIG. 3A shows the case where the time delay Tc = 0 at which the value of the correlation function becomes maximum, FIG. 3B shows the time delay Tc> 0 when the value of the correlation function becomes maximum, and FIG. This is the case where the time delay Tc <0 at which the value of the correlation function becomes maximum. In each case, the first signal is multiplied by a window function whose amplitude gradually increases with time, and the second signal is multiplied by a window function whose amplitude gradually decreases with time. Add only after delaying Tc. Here, the shape of the window function is changed based on the time delay Tc at which the correlation function becomes maximum. The time length of the addition result is (T-Tc).

FIG. 4 schematically shows a processing example of the above-described voice speed conversion method of the present invention.

FIG. 4A shows an input signal, and FIG. 4B shows an output signal when the time axis conversion ratio α = 3/2. The time delay T _c1 = 0 at which the correlation function of X _A1 and X _B1 becomes maximum, the time delay T _c2 > 0 at which the correlation function of X _A2 and X _B2 becomes maximum, and the correlation function of X _A3 and X _B3 become maximum. The time delay T _c3 <0. The sum of the time lengths of the signal obtained by adding the first signal X _An and the second signal X _Bn and the third signal X _Cn following the first signal X _An is α (T−T _cn ) / ( α-1), which is determined based on the time-axis conversion ratio α, the time delay T _{cn at} which the value of the correlation function is maximum, and the finite time length T. Input signal (X _C1 + X
The ratio of the time length of the output signal to the time length of ( _C2 + X _C3 ) is
It becomes equal to the set time axis conversion ratio α (= 3/2). X
_{Since Cn} is output as it is and all the sections of the input signal are used, there is no information loss in the output signal.

As described above, according to the present embodiment, the first signal X _A is multiplied by the gradually increasing window function, and the second signal X _B is multiplied by the gradually decreasing window function, and then the addition is performed, thereby performing the addition. The discontinuity of the signal amplitude is reduced. Then, the first signal multiplied by the window function and the second signal multiplied by the window function are added at the position of the time delay Tc at which the value of the correlation function becomes maximum, whereby the discontinuity of the phase is reduced. Further, a signal obtained by adding the first signal multiplied by the window function and the second signal multiplied by the window function and the third signal X _C subsequent to the first signal X _A are correlated with the time axis conversion ratio α. By outputting only the time length determined based on the time delay Tc that maximizes the function value and the finite time length T, the input signal can be easily input without time loss and the time axis conversion ratio α can be set to α ≧ 1.0. The output can be expanded to the range of.

A second embodiment of the present invention will be described below with reference to the drawings. The present invention is a voice speed at which a time-axis conversion ratio α can be output within a range of α ≦ 1.0, which is a voice with few discontinuities in the amplitude and phase of the signal and which does not cause data loss. It provides a conversion method.

FIG. 5 shows a flow chart of a voice speed converting method in the second embodiment of the present invention. The operation will be described below.

First, in step 51, the input pointer is reset. Next, in step 52, the first signal X _{A in} the T section is input from the input pointer. Then, in step 53, T is added to the input pointer.
Next, at step 54, the second signal X _{B in} the T section is input from the input pointer. Then, in step 55, the correlation function of the first signal X _A and the second signal X _B is calculated, and the time delay Tc at which the value of this correlation function becomes maximum is searched for.

Next, at step 56, the first signal X is calculated based on the time delay Tc at which the correlation function obtained above becomes maximum.
Multiply _A by a decreasing window function. And in step 57,
The second signal X _B is multiplied by the gradually increasing window function on the basis of the time delay Tc at which the correlation function obtained above becomes maximum. Next, at step 58, the first signal multiplied by the window function and the second signal multiplied by the window function are time-delayed T with which the correlation function becomes maximum.
Add after shifting to the position of c. And step 59
Then, add T to the input pointer. Next, in step 60, the signal added in step 58 and the second signal X _B
A signal subsequent to, that is, a third signal X _C having the current input pointer as a starting point is output only in the α (T-Tc) / (1-α) section. Then, in step 61, (2αT-T-Tc) / (1-α) is added to the input pointer. After this,
Return to step 52.

The process of calculating the correlation function of the first signal X _A and the second signal X _B in step 55 of FIG. 5 and searching for the time delay Tc at which the value of the correlation function is maximum is shown in FIG. This is the same as in the first embodiment of the present invention shown.

FIG. 6 is a schematic diagram of the processing in steps 56, 57 and 58 of FIG.

FIG. 6A shows the case where the time delay Tc = 0 at which the value of the correlation function becomes maximum, FIG. 6B shows the case where the time delay Tc> 0 at which the value of the correlation function becomes maximum, and FIG. This is the case where the time delay Tc <0 at which the value of the correlation function becomes maximum. In each case, the first signal is multiplied by a window function whose amplitude gradually decreases, and the second signal is multiplied by a window function whose amplitude gradually increases with time. Add only after delaying Tc. Here, the shape of the window function is changed based on the time delay Tc that maximizes the correlation function. The time length of the addition result is (T + Tc).

FIG. 7 schematically shows a processing example of the voice speed conversion method described above. FIG. 7A shows an input signal,
(B) is an output signal when the time-axis conversion ratio α = 2/3. Time delay T _c1 = where the correlation function of X _A1 and X _B1 is maximum =
0, time delay T _c2 > at which the correlation function of X _A2 and X _B2 becomes maximum>
0, and the time delay T _c3 <0 at which the correlation function of X _A3 and X _B3 becomes maximum. First signal X _An and second signal X
The sum of the time lengths of the signal obtained by adding _Bn and the third signal X _Cn that follows the second signal X _Bn is α (T-T _cn ) / (1-α), and the time-axis conversion ratio α and It is determined based on the time delay T _cn that maximizes the value of the correlation function and the finite time length T. The ratio of the time length of the output signal to the time length of the input signal is equal to the set time axis conversion ratio α (= 2/3). Since the input signals are all used for the first signal X _An , the second signal X _Bn , and the third signal X _Cn , there is little information loss in the output signal.

As described above, according to the present embodiment, the first signal X _A is multiplied by the gradually decreasing window function, and the second signal X _B is multiplied by the gradually increasing window function, and then the addition is performed, thereby performing the addition. The discontinuity of the signal amplitude is reduced. Then, the first signal multiplied by the window function and the second signal multiplied by the window function are added at the position of the time delay Tc at which the value of the correlation function becomes maximum, whereby the discontinuity of the phase is reduced. Further, a signal obtained by adding the first signal multiplied by the window function and the second signal multiplied by the window function and the third signal X _C subsequent to the second signal X _B are added to the time axis conversion ratio α and the correlation function. By outputting only the time length determined based on the time delay Tc that maximizes the value of T and the finite time length T, the input signal can be easily reduced with a time axis conversion ratio α of α ≦ 1.0. It can be compressed into a range and output.

[0042]

As is apparent from the above description, according to the present invention, the first signal and the second signal are multiplied by a window function whose amplitudes change in a complementary manner with respect to time, and then the signals are added. By adding the first signal multiplied by the window function and the second signal multiplied by the window function at the time delay position where the value of the correlation function becomes maximum, , Phase discontinuity is reduced.

Furthermore, a signal obtained by adding the first signal multiplied by the window function and the second signal multiplied by the window function and the third signal following the added signal are added to the time axis conversion ratio α and the correlation function. By outputting only the time length determined based on the time delay Tc that maximizes the value of and the finite time length T, it is possible to obtain an excellent effect that there is little loss of signal and conversion to an arbitrary speed can be performed. be able to.

[Brief description of drawings]

FIG. 1 is a flowchart of a voice speed conversion method according to a first embodiment of the present invention.

FIG. 2 is a flowchart of a correlation function calculation of the voice speed conversion method according to the first embodiment of the present invention.

FIG. 3 is a schematic diagram of weighting by a window function and addition at a time delay position where the value of the correlation function is maximum in the voice speed conversion method according to the first embodiment of the present invention.

FIG. 4 is a schematic diagram of an input signal and an output signal of the voice speed conversion method according to the first embodiment of the present invention.

FIG. 5 is a flowchart of a voice speed conversion method according to the second embodiment of the present invention.

FIG. 6 is a schematic diagram of weighting by a window function and addition at a time delay position where the value of the correlation function is maximum in the voice speed conversion method according to the second embodiment of the present invention.

FIG. 7 is a schematic diagram of an input signal and an output signal of a voice speed conversion method according to a second embodiment of the present invention.

FIG. 8 is a block diagram of a conventional voice speed conversion device.

FIG. 9 is a schematic diagram of an input signal and an output signal of a conventional voice speed conversion device.

[Explanation of symbols]

 81 A / D converter 82 Buffer 83 Speed control circuit 84 Data reading circuit 85 Muting circuit 86 D / A converter

Claims (3)

[Claims] 1. A time delay in which a correlation function between a first signal having a finite time length T and a second signal having a finite time length T following the first signal is calculated to maximize the value of the correlation function. Tc is obtained, and each of the first signal and the second signal is multiplied by a window function whose amplitude temporally changes complementarily, which is determined based on a time delay Tc at which the value of the correlation function is maximum, and The first signal multiplied by the window function and the second signal multiplied by the window function are added at the position of the time delay Tc where the value of the correlation function is maximum, and the third signal is added to the added signal. Based on the time-axis conversion ratio α (= output time / input time), the time delay Tc at which the value of the correlation function is maximum, and the finite time length T, which are continuously output and the added signal and the third signal are output. Output for the determined time length, and the first signal and the second signal in the next processing
The starting point of the signal is determined based on the time-axis conversion ratio α, the time delay Tc at which the value of the correlation function is maximum, and the finite time length T,
A voice speed conversion method, characterized in that the reproduction time of voice is changed with respect to the length of an original sound by repeating all the processes described above. 2. A time delay Tc at which the correlation function has a maximum value by calculating a correlation function between a first signal having a finite time length T and a second signal having a finite time length T following the first signal. And the first signal is multiplied by a window function whose amplitude gradually increases with time, which is determined based on the time delay Tc at which the value of the correlation function becomes maximum, and the second signal has the correlation function of the correlation function. The first signal multiplied by the window function and the second signal multiplied by the window function are multiplied by a window function in which the amplitude determined based on the time delay Tc having the maximum value is gradually reduced. Addition is performed at the position of the time delay Tc where the value of the function is maximum, the third signal is continuously output to the added signal, and the added signal and the third signal are time length {α (T- Tc) / (α-
1)} is output (where α is the time base conversion ratio,
Output time / input time), the starting point of the first signal in the next processing is a point obtained by delaying the starting point of the first signal by {(T-Tc) / (α-1)}. ,
A voice speed conversion method characterized in that the reproduction time of voice is changed to 1.0 times or more of the length of the original sound by repeating all the processes described above. 3. A time delay Tc at which a correlation function has a maximum value by calculating a correlation function between a first signal having a finite time length T and a second signal having a finite time length T following the first signal. Is obtained, and the first signal is multiplied by a window function in which the amplitude determined based on the time delay Tc at which the value of the correlation function is maximum gradually decreases, and the second signal is calculated by the correlation function of the correlation function. The correlation is made between the first signal multiplied by the window function and the second signal multiplied by the window function, which is obtained by multiplying the window function by which the amplitude determined based on the time delay Tc having the maximum value gradually increases with time. Addition is performed at the position of the time delay Tc where the value of the function is maximum, the third signal is continuously output to the added signal, and the added signal and the third signal are time length {α (T- Tc) / (1-
α)} only is output (where α is the time base conversion ratio,
Output time / input time), and the starting point of the first signal in the next processing is a point obtained by delaying the starting point of the first signal by {(T-Tc) / (1-α)}. ,
A voice speed conversion method characterized in that the reproduction time of a voice is changed to 1.0 times or less of the length of an original sound by repeating all the processes described above.