JP3156020B2 - Audio speed conversion method - Google Patents

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 for changing only the duration without changing the fundamental frequency of voice.

[0002]

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

[0003] Hereinafter, the above-described conventional audio speed converter will be described with reference to the drawings.

FIG. 8 shows the configuration of a conventional voice speed converter. In FIG. 8, reference numeral 81 denotes an AD converter, 82
Is a buffer, 83 is a speed control circuit, 84 is a data read circuit, 85 is a muting circuit, and 86 is a DA converter.

[0005] The operation of the audio speed converter having the above configuration will be described below.

First, an input signal is converted into a digital signal by an AD converter 81 and written into a buffer 82. Next, the speed control circuit 83 controls the data read circuit 84 in accordance with the compression / expansion ratio to read data from the buffer 82. Depending on the data reading method, the reproduction speed can be variously changed. To shorten the reproduction time, the data to be read is thinned out in block units. To lengthen the reproduction time, the data to be read is repeated in block units. The discontinuous portion between the blocks is muted by a muting circuit 85, converted into an analog signal by a DA converter 86, and output.

FIG. 9 shows that the compression / expansion ratio (time axis compression / expansion ratio = the ratio of the time length of the output signal to the input signal) α is 0.5 and 2.0.
Is schematically shown. (A) shows the case of the original sound, (b) shows the case where the time base conversion ratio is 0.5, and (c) shows the case where the time base conversion ratio is 2.0.

[0008]

However, in the above-described conventional configuration, when the time axis is compressed to increase the speed,
In order to reduce data, consonants are missing and the clarity is reduced, and the connection points of the blocks are discontinuous.To reduce this, the connection points are muted, but the amplitude and phase are discontinuous. There was a problem that only a voice with poor naturalness could be obtained.

[0009] In another conventional voice speed converter,
TDHS (Time Domeim Harmonic
Although there is a method of using the pitch period of the input signal as in Scaling, it is not applicable when music or noise is superimposed on the input signal because it is difficult to extract the pitch. In addition, the window length for weighting and adding the waveform changes depending on the speed ratio and the pitch period. When a signal including a signal that is lower than the obtained pitch frequency is weighted and added, the low-frequency components tend to be connected discontinuously. And has a problem of lack of smoothness.

The present invention has been made in view of the above problems, and has low discontinuity in both amplitude and phase of a waveform, can output natural sound without causing data loss, and can output music signals and the like. An object is to smoothly reproduce a signal including a low frequency component.

[0011]

The present invention according to claim 1 is provided.
Represents a signal having a predetermined time length Ts as A,
When a signal having a time length Ts subsequent to the signal A is B.
And the time length that is a time delay k (0 ≦ k) with respect to the signal A
A time delay -k (0 <
k), a signal B ′ having a time length Ts, and a signal A and a signal
Correlation function between signal B 'and signal A' and signal B
The number is calculated within a predetermined range of k to maximize the correlation function.
The time delay rk is determined, and rk is determined according to the value of rk.
When = 0, the signal A and the signal B are gradually reduced by the width of the time length Ts.
Weighted addition is performed in the relation of gradual increase, and rk> 0
In the case of, after outputting the signal A with the time width rk, the signal A '
Signal B is weighted in a relationship of gradual decrease and increase in the width of time length Ts
The signal A and the signal are output when rk <0.
B 'is weighted in a relationship of gradual decrease and increase with the width of the time length Ts.
And output it, and after the process for the value of rk,
Axis compression / expansion ratio (time length of output signal with respect to input signal
Ratio) α and the time delay rk, the expression ｛α (Ts−
rk) / (1−α)｝ until the given time length is reached.
A series of processes for outputting a signal subsequent to the addition signal is performed as follows.
The expression {(Ts-rk) / (1-α)} gives the head of the signal A.
To reset to the point delayed by the length of time
Change the audio playback time to less than 1.0 times the original sound.
This is an audio speed conversion method.

Further, the present invention according to claim 2 provides a voice signal.
A signal having a predetermined time length Ts is denoted by A, and the signal A
When a signal having a time length Ts subsequent to
On the other hand, a signal having a time length Ts with a time delay k (0 ≦ k)
A ′ and a time delay −k (0 <k) with respect to the signal B.
For the signal B 'having the time length Ts, the signal A and the signal B'
The correlation function and the correlation function between signal A ′ and signal B
Time delay when the correlation function is maximized when calculated in the range of k
rk is obtained, and in the case of rk = 0,
In this case, the signal B and the signal A are gradually reduced by the width of the time length Ts.
And weighted and added, and if rk <0,
After outputting the signal B with a time width (-rk), the signal B ′ is output.
Weighted with signal A in the relationship of gradually decreasing and increasing with the width of time length Ts
Calculated and output, also, rk> 0, the signal B and the signal
A 'is weighted in the relationship of gradual decrease and increase with the width of the time length Ts.
And output it, and after the process for the value of rk,
Axis compression / expansion ratio (time length of output signal with respect to input signal
Ratio) α and the time delay rk, the expression ｛α (Ts−
rk) / (α-1)} until the given time length is reached.
A series of processes for outputting a signal subsequent to the addition signal is performed as follows.
The expression {(Ts-rk) / (α-1)} gives the head of the signal A.
To reset to the point delayed by the length of time
More than 1.0 times the original sound
This is an audio speed conversion method.

[0013]

According to this configuration, the signal A 'and the signal B or the signal A and the signal B' are weighted and added with the width of the time length Ts, thereby reducing the loss of the added signal and the discontinuity of the amplitude. Since the weighted addition is performed with a fixed time length Ts, it is possible to smoothly connect signals including low frequency components. Further, a time delay rk at which the correlation function between the signal A ′ and the signal B or the signal A and the signal B ′ is maximized.
, The phase mismatch before and after the section where the waveform connection is performed is reduced.

[0014]

EXAMPLES Hereinafter, a first embodiment of the present invention will be described with reference to the drawings.

According to the present invention, the companding ratio α is determined by the formula ΔTs + kmax /
The present invention relates to an audio speed conversion method operating in the range of 2 · Ts｝ ≦ α ≦ 1.0.

FIG. 1 is a flow chart of a voice speed conversion method according to a first embodiment of the present invention, and its operation will be described.

In this example, the audio signal is discrete time data x
It is assumed that sampling is performed at (n). In the following processing, data is specified using P1, P2 as an input data pointer and an output data pointer P3.
First, in step 101, an address ip1 indicated by the input pointer P1 is set to the head address of audio data to be reproduced. The address ip2 indicated by P2 indicates the data Ts times after P1. Also, an initial value is set to the address op indicated by the output pointer.
In step 102, the companding ratio α is set. The companding ratio α satisfies the value shown in the above equation.

Next, one of the signal A having the number of data Ts from the pointer P1 and the signal B having the number of data Ts from the pointer P2 is shifted with respect to one of the signals A in a time delay direction, and the position where the correlation becomes high is determined. Step 10 to find
In step 3, a correlation function is calculated, and in step 104, the number of data (time delay) rk corresponding to a time delay when the correlation function is maximized is obtained. Range of audio data to be used according to the positive or negative value of the time between the delay k as shown in FIG. 2 is different from the calculation contents of the correlation function COR. The range of the time delay k for performing the calculation is a maximum value kmax and a minimum value kmin
Is set in advance, and the range for obtaining the correlation delay is limited. The time delay rk at which the correlation function is maximized is obtained as described above. In step 105, the number of data Tt for directly outputting audio data is calculated as shown in FIG. The calculation of the number of data Tt in this straight-out section is also delayed by rk.
The calculation formula differs depending on the sign of.

If the value of the time delay rk is positive, the processing of steps 107, 108 and 109 is performed to obtain an output waveform, and otherwise, the processing of steps 110 and 111 is performed to obtain an output waveform. . Here, step 10
The Wdec (i) at 8, 110 is linearly monotonically decreasing with the increase of i when i is 0 and i is ( T).
s-1 ) is a window function that becomes 0 at the time of (s-1 ) . In addition, Winc (i) in steps 108 and 110 is linearly monotonically increased at 0 when i is 0 and increases as i increases, and i becomes (T
The window function becomes 1 in the case of s-1).

The values in FIG. 4 between two o'clock delay r k is 0, a positive, the output waveform divided for negative shows how sought. It can be seen that the number of data Tt is shorter when the time delay rk is positive than when the time delay rk is zero. Conversely, when the time delay rk is negative, the data number Tt is long. This is to adjust the length of the number of data Tt according to the deviation of the time delay rk so that there is no deviation from the target companding ratio α. If the processing is to be continued, the address pointed to by the input data pointer and the address pointed to by the output data pointer is updated as shown in step 113, and then the processing from step 102 onward is repeated.

As described above, according to the present embodiment, it is possible to realize a method of listening by compressing the reproduction time having the following features (a method of increasing the speed without changing the pitch). it can. A correlation function based on the pointers P1 and P2 is calculated, and weighted addition is performed at a position where the correlation is high. This prevents the phase from becoming significantly mismatched in the section before and after connecting the waveforms. One of the signals at the two separated portions is monotonically decreased, and the other is added after applying a monotonically increasing window function, so that the continuity of the amplitude in the section where the waveforms are connected is maintained well.

As a result, it is possible to obtain a clear reproduced sound that is smooth and natural, and has a clear information loss and little echo feeling. Further, the number of data in the straight-out section following the weighted addition is calculated after the number of time-delayed data rk is determined, and a change in the companding ratio α due to a change in the number of time-delayed data may not occur. Absent. Further, the length of the section to be weighted and added is such that the waveform is cross-fade with a constant length Ts irrespective of the input signal and the time delay rk, so that the cross-fade length is not shortened by the value of the time delay rk. Thus, a smooth reproduced sound of low frequency components contained in the connected signal can be obtained.

Hereinafter , a second embodiment of the present invention will be described with reference to the drawings. In the present invention, the drawing / drawing ratio α is represented by the formula 1.
An object of the present invention is to provide a voice speed conversion method that operates in the range of 0 ≦ α ≦ (Ts / kmax) .

FIG. 5 is a flowchart of a voice speed conversion method according to a second embodiment of the present invention, and the operation will be described.

In this example, as in the first embodiment, the audio signal is sampled as discrete-time data x (n), and data is designated using input data pointers P1, P2 and output data pointer P3. Do. First, in step 501, the address ip1 indicated by the input pointer P1
Is set to the head address of the audio data to be reproduced. Also, the address ip2 indicated by P2 has a value from P1 to T.
Point to data s times later. Also, an initial value is set to the address op indicated by the output pointer. Step 5
In 02, the companding ratio α is set. The companding ratio α satisfies the value shown in the above equation in the second embodiment. Next, one of the signal A having the number of data Ts from the pointer P1 and the signal B having the number of data Ts from the pointer P2 is shifted with respect to one of them in a time delay direction to obtain a position having a high correlation. In step 503, a correlation function is calculated, and in step 504, the number of data rk corresponding to a time delay when the correlation function is maximized is obtained. Correlation function C
The calculation contents of OR are calculated as shown in FIG. 2 as in the first embodiment.

Further, the range of calculated line cormorants time between delay k is previously set maximum value kmax and the minimum value kmin advance, in the range correlating delay to limit. The time delay rk at which the correlation function is maximized is obtained as described above. In step 505, the number of data Tt for directly outputting audio data is calculated as shown in FIG. The calculation formula of the number Tt of data in the straight-out section also differs according to the sign of the time delay rk. When the value of the time delay rk is negative, the processing of steps 507, 508, and 509 is performed to obtain an output waveform. Otherwise, the processing of steps 510 and 511 is performed to obtain an output waveform. Here, step 508,
The window Wdec (i) at 510 is a window having a magnitude of 1 when i is 0, linearly decreasing monotonically with the increase of i, and becoming 0 when i is ( Ts−1 ) , as in the first embodiment. Function. Also, Winc in steps 508 and 510
(I) is 0 and i is 0 when i is 0 as in the first embodiment.
Is a window function that linearly and monotonically increases as i increases and becomes 1 when i is ( Ts−1 ) .

The values in FIG. 7 between two o'clock delay r k is 0, negative, positive divided by the output waveform in the case shows how sought. It can be seen that the number of data Tt is shorter when the time delay rk is positive than when the time delay rk is zero. Conversely, when the time delay rk is negative, the data number Tt is long. This is to adjust the length of the data number Tt according to the deviation of the time delay rk so that there is no deviation from the target companding ratio α. If the processing is to be continued, the address pointed to by the input data pointer and the address pointed to by the output data pointer are updated as shown in step 513, and then the processing from step 502 onward is repeated.

As described above, according to the present embodiment, it is possible to realize a method of extending the reproduction time and having the following characteristics for listening (a method of reducing the speed without changing the pitch). it can. A correlation function based on the pointers P1 and P2 is calculated, and weighted addition is performed at a position where the correlation is high. This prevents the phase from becoming significantly mismatched in the section before and after connecting the waveforms. One of the two separated signals is monotonically reduced, and the other is multiplied by a monotonically increasing window function, and then added, so that the continuity of the amplitude in the section where the waveforms are connected is maintained well. As a result, it is possible to obtain a clear and natural reproduced sound which is unprecedented and smooth and has little information loss and echo feeling.

The number of data in the straight-out section following the weighted addition is the number of data rk with a time delay.
Is calculated after the determination is made, and there is no shift in the companding ratio α due to a change in the number of data rk with a time delay. Furthermore, the length of the section to be weighted and added is such that the cross-fade length is not shortened by the value of the time delay rk because the waveform is connected by cross-fading with a constant length Ts irrespective of the input signal and the time delay rk. Thus, a smooth reproduced sound of low frequency components contained in the connected signal can be obtained.

[0030]

According to the present invention, the time delay rk that maximizes the correlation function based on one of the signal A and the signal B is determined, and the position where the waveform is weighted and added is changed according to the time delay. The phase is prevented from being significantly mismatched before and after the section where the signal is connected. In addition, since the signal is multiplied by a window function that gradually decreases in time and a window function that gradually increases in time in the section where the waveform is connected, the discontinuity of the amplitude in the section where the waveform is connected is eliminated. .
Further, after determining the time delay rk, the equation ｛α (Ts−
rk) / (11−α)} or the formula {α (Ts−rk) /
Until the output time length reaches the value shown in (α-1)｝ ,
Since the signal subsequent to the weighted and added signal is output as it is, there is no deviation from the companding ratio α due to a change in the number of time delay data. Further, by performing the weighted addition with the width of the fixed time length Ts, there is an advantage that a smooth reproduced sound of low frequency components included in the connected signal can be obtained.

[Brief description of the 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 operation of the voice speed conversion method according to the first embodiment of the present invention.

FIG. 3 is a flowchart for calculating a length of a straight-out section according to the first embodiment of the present invention;

FIG. 4 is a schematic diagram of an output signal obtained by weighting and adding an input signal by a value of a time delay rk in 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 a second embodiment of the present invention;

FIG. 6 is a flowchart for calculating the length of a straight-out section according to the second embodiment of the present invention;

FIG. 7 is a schematic diagram of an output signal obtained by weighting and adding an input signal by a value of a time delay rk in the audio speed conversion method according to the second embodiment of the present invention.

FIG. 8 is a configuration 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 audio speed conversion device.

[Explanation of symbols]

 A, B signal Ts Predetermined time length rk Time delay α Companding ratio

Continuation of front page (56) References JP-A-3-219462 (JP, A) JP-A-4-104200 (JP, A) JP-A-4-188199 (JP, A) JP-A-6-222794 (JP) , A) (58) Fields studied (Int. Cl. ⁷ , DB name) G10L 11/00-11/06 G10L 21/00-21/04

Claims (2)

(57) [Claims] 1. An audio signal having a predetermined time length Ts
A signal is represented by A, and a signal having a time length Ts subsequent to the signal A is represented by B.
Is a time delay k (0 ≦ k) with respect to the signal A.
Time delay with respect to the signal A 'having a time length Ts
k (0 <k), a signal B ′ having a time length Ts
The correlation function between signal A and signal B 'and signal A' and signal B
Is calculated in a predetermined range of k, and the correlation function is
The maximum time delay rk is determined, and corresponding to this rk value,
Therefore, when rk = 0, the signal A and the signal B are separated by a time length Ts.
Weighted addition is output in the relationship of gradually decreasing and increasing in width, and
If rk> 0, signal A is output with time width rk and then
The signal A 'and the signal B are gradually reduced and increased in the width of the time length Ts.
Weighted addition is performed, and when rk <0, the signal is
A and the signal B ′ overlap in a gradually decreasing and increasing relationship with the width of the time length Ts.
And outputs the result after the process for the value of rk.
The time axis compression / expansion ratio (for the output signal with respect to the input signal,
｛Α corresponding to the ratio (interval length) α and the time delay rk
Reaches the time length given by (Ts-rk) / (1-α)｝
A series of processes for outputting a signal subsequent to the addition signal up to
Is calculated by adding the head of the next signal A to the equation ｛(Ts−rk) / (1−
α) Reset to the point delayed by the time length given by｝ and repeat
By returning, the playback time of the sound is less than 1.0 times the original sound
Voice speed conversion method that is changed to. 2. An audio signal having a predetermined time length Ts
A signal is represented by A, and a signal having a time length Ts subsequent to the signal A is represented by B.
Is a time delay k (0 ≦ k) with respect to the signal A.
Time delay with respect to the signal A 'having a time length Ts
k (0 <k), a signal B ′ having a time length Ts
The correlation function between signal A and signal B 'and signal A' and signal B
Is calculated in a predetermined range of k, and the correlation function is
The maximum time delay rk is determined, and corresponding to this rk value,
Therefore, when rk = 0, the signal B and the signal A are separated by a time length Ts.
Weighted addition is output in the relationship of gradually decreasing and increasing in width, and
When rk <0, the signal B was output with a time width (-rk)
Thereafter, the signal B 'and the signal A are gradually reduced and increased by the width of the time length Ts.
Weighted and added in relation to each other, and when rk> 0,
In this case, the signal B and the signal A ′ are gradually reduced by the width of the time length Ts.
Weighted and added in relation to output
After processing, the time axis compression / expansion ratio (output for input signal
Signal time ratio) α and the time delay rk
The time length given by the equation {α (Ts−rk) / (α−1)}
A series of signals that follow the sum signal until they reach
Is calculated by adding the head of the next signal A to the equation ｛(Ts−rk) /
(Α-1) is reset to the point delayed by the time length given by
To repeat the playback time of the original sound by 1.0
An audio speed conversion method that changes it more than twice.