JPS60140298A - Speech controller - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a speech control device which allows, for example, the pitch and speed of an audio signal to be changed arbitrarily.

Background Art and Problems There is a speech control device that allows the speed and pitch of an audio signal to be changed arbitrarily. Conventionally, in such a device, the speed and pitch of a signal are changed by operating on the time axis as follows. For example, if there is a signal as shown in Figure 1A, in order to increase the speed of this signal, the signal is extracted intermittently according to the rate of increase, and the extracted signal is as shown in Figure 1B. Connect and output. It is also possible to roughly extend the time axis of this signal as shown in FIG. 1C to lower the pitch.

However, in the case of such devices, discontinuities occur at the joints of signals, and although countermeasures such as crossfading have been taken, it is still not possible to completely eliminate discontinuities, and the auditory sense is affected. There was a feeling of discomfort above.

Additionally, the information that was not retrieved intermittently was lost, making the signal incomplete.

In view of this, the inventor of the present application has previously proposed a speech control device that does not have discontinuities or dropouts in signals.

In Fig. 2, the audio signal supplied to the input terminal (1) is supplied to the AD conversion circuit (2) and the digital signal x
cm> and stored in the buffer memory (3). This memo (January 3), for example, has an overall length of L, and is stored by sequentially shifting the human power. Every time the contents of this memory (3) are R-shifted (R<L), the memory (3) is
The contents of are taken out in parallel to the buffer memory (4).

As a result, signals are extracted from the memory (4) at arbitrary time intervals R and for arbitrary lengths of time.

Here, R is made sufficiently small with respect to L, and each signal is overlamped with respect to each other.

The signal from this memo (January 4) is supplied to the multiplier (5) and a predetermined window coefficient hQIl) is converted to H(H).The signal obtained by B(of this window coefficient) is supplied to the Fourier transform circuit (6). .

As a result, the time axis of the 'ζ signal is converted to the frequency axis.

This converted signal is supplied to a multiplier (7) to perform phase adjustment corresponding to the shift amount R in Memo January 3). This phase-adjusted signal is supplied to a processing circuit (8).

- In this processing circuit (8), the signal converted into the frequency axis by Fourier transform is stored in each memory address for each predetermined frequency range (1). The stored signals are sequentially read out.

The signal output from this processing circuit (8) is sent to the multiplier (9).
), and the phase adjustment corresponding to the shift amount R' at the time of output, which will be described later, is performed. This phase-adjusted signal is supplied to an inverse Fourier transform circuit (10).

This converts the frequency axis of the signal into the time axis.

This converted signal is supplied to a multiplier (11) and multiplied by a window coefficient f(2) corresponding to the above-mentioned window coefficient h(T11). The signal multiplied by this window coefficient is stored in the buffer memory (12). The contents of this memory (12) are supplied in parallel to the buffer memory (13).

For example, the total length of this memory (13) is L,
The contents are sequentially shifted and output. Further, O is stored in the blank space created by the shift. Each time the contents of this memory (13) are shifted by R', the contents of the memory (12) are supplied and added to the previous contents.

The signal from this memory (13) is
) is supplied to the output terminal (15) and converted into an analog signal.
It is taken out.

Furthermore, the signal from the input terminal (1) is supplied to the /S Ibus filter (21) and the low-pass filter (22). These outputs are supplied to a comparator circuit (23), and the energies of the signals in each band are compared. This comparison output is supplied to a selection circuit (24) for window coefficients I1 and f(II+), and a window coefficient corresponding to each case of ζ is selected.

In this device, when a signal as shown in FIG. 3A is supplied to the input terminal +11, this signal is extracted for each time interval R by one time length. This extracted signal is Fourier transformed' (as shown in FIG. 3B, a spectrum in which the time axis is transformed to the frequency axis is formed).

This signal is stored in each memory address of the processing circuit (8) and phase adjusted. The i1'i, 411 i11+I adjusted I symbol is subjected to inverse Fourier transform to form signals sequentially shifted by time intervals R as shown in FIGS. 4A and 5A. These signals are sequentially supplied to memory (13) through memory (12) and added.

Here, for example, when increasing the speed of the signal, the shift l-amount R' when the signals C in the memory (13) and the signals from the memory (12) are added is set to (R'< R). When addition is performed for each shift acid R°, this signal becomes as shown in FIG. 4B, R' The frequency band of this signal remains the same, but the time is shortened to □.

Furthermore, when the speed is to be lowered, R'>R.

As a result, the added signal becomes a signal whose frequency band R'' remains the same as shown in FIG.

R.

Furthermore, by extracting this signal with the clock of □,
The speeds of the signals are respectively returned to the same level, and a signal whose frequency band is lowered or increased to the R' frequency band or □ is obtained as shown in FIG. 4C and FIG. 5C.

Although the waveforms in FIGS. 3 to 5 are shown in analog form, they are actually processed using digital values.

Furthermore, in the above device, window coefficients hQ11), f(I
II) is made into the following relationship. That is, signal XQ
I+), ζ xgn)-+h (SR-m)
X(m)e-j''II+m-1F.Furthermore, inverse Fourier transform is performed to obtain ζS=-1.Since it is sufficient that this is equal to X(T11), ψ1 is sufficient.

Then, as described above, when the spectral shape of the input signal is detected and the window coefficients h(tl) and f(III) are selected, for example, when the low frequency component is smaller, h(Tl
l)"1 fcm>= 0.5-0.5 cos (2πn/
N-1) n=Q,...N-1 When the low frequency component is larger, han+=0.54-0.46 cos (2tc n
/N −1) n=0, . . . N−1 f(Wo=2π/Σhj) The sound quality can be improved each time.

Further, in the above-mentioned device h', the phase adjustment in the multiplier (9) is performed as follows.

First, the spectrum after Fourier transformation at time SR is X (St?, ωk), its real part is XR (SR, ωk), its imaginary part is X, I (SR, ωk), and the single value of the phase is P (Si2.ωk). , -π≦P (SR, ωk)<π and the phase made continuous along the time point S is p (SR, ωk), where -■<'p, (S)l, ωk). At this time, phase continuity and phase transformation are performed as follows.

i) If S≠0 (al First, by Fourier transform, X (Si2.
ωk).

(Creation) Next, add 1' (311, ωk).

When the sign of XR (SR, ωk) and Xr (SR, ωk) is (+, +) or (+, -), P (SR+
ωk) = jan-” (XI (SR, ωk)/
XR (SR, ωi +)) When the sign is (-, +), P (SR, (dk, ) = tan-" (XI (
SR, ωk)/XR(SR, ωk)) + π When the sign is (-, -), P (SR, ωk) = tan-" (XI(St?
, ωlO/XR(SR, ωk))−π.

101 Furthermore, I P (SR, ωk')-P((S-1) R, ω
k) 1<ε However, it is determined whether ε is a constant.

(di And when this stops, P (SR, (ilk) = p ((S-1)R, ωk)
+P C5R, ωk) - P((S-1) R, ωk).

(d') Also, if (C) is incorrect, first, when the sign of the absolute value is (-), p (SR, ωk) = p ((S-1) +5. ωk)
+P (St?, ωk)-P((S-1) R, ωk
)+2π When the sign is (+), p (SR, ωk) = p((S-1) R, ωk)
+P (SILωk)2P((S-1) R, ωk)-
Let it be 2π.

As described above, the phase is made continuous.Furthermore, when changing the shift amount during synthesis as described above, in order to prevent interference in encoding and decoding, the shift amount must be changed. According to the change, Rp (SR, ωk)→p(S15.ωk) -. This prevents noise from occurring due to phase discontinuity.

This is how speech control is performed, but with this device, the signal converted to the frequency axis by Fourier transform is phase-aligned and then synthesized by G, so an extremely high quality signal can be obtained. This provides good speech control without signal discontinuities or dropouts.

By the way, in this device, in order to avoid a change in sound quality between the time length extracted and the shift amount R, for example, when a Hamming line is used as the extraction window coefficient, R< ---There is a restriction that it is L. For this reason, if you try to take small units of change in speech control,
A problem arises in that L becomes extremely large.

In other words, for example, if a signal sampled at 10 kHz is windowed at L = 128 points to make one frame, and this is shifted by R - 32 points for extraction, then R' = 31 points. By adjusting the °ζ phase 1 and performing synthesis, a signal 2 whose time has been shortened to - is extracted. This gives a - change in time of 2.

However, if the unit of this change is made smaller, for example, it becomes necessary, and the extraction time length becomes 4 L.

When L becomes large in this way, there are problems such as an increase in the amount of calculation during Fourier transform, an increase in processing time, and an increase in the size of hardware for processing.

OBJECTS OF THE INVENTION In view of these points, the present invention makes it possible to finely refine the (14th) position of change without increasing the amount of calculations.

SUMMARY OF THE INVENTION The present invention provides means for extracting human-generated audio signals at arbitrary time intervals and arbitrary lengths of time, and means for performing Fourier transform on each extracted frame to convert the time axis into the frequency axis. , a means for adjusting the phase of the converted signal No. 1, a means for performing an inverse Fourier transform on this phase IM adjusted signal to inversely transform the frequency axis into a time axis, and a means for inversely converting the inversely transformed signal into a predetermined value. C1 is a speech control device comprising means for interpolating at a magnification of Accordingly, the unit of change can be made finer without increasing the amount of calculation.

In FIG. 6 of the embodiment, °ζ, multiplier (11) after inverse Fourier transform
An interpolation circuit (31) is provided after. Then, the interpolation circuit (31) interpolates the "ζ" and "Zuba" numbers, for example, by four times.

For example, when the G8 signal for each sampling period as shown in FIG. 7A is input from the multiplier (11), three points are generated between each signal as shown in FIG. 7B. performs interpolation. The double circle indicates Noji No. 14.

As a result, in the synthesis using memories (12) and (13), the 4R points can be shifted one point at a time, as shown in FIG. 7C. In the conventional Zunawaji system, it was possible to shift only to the position of the original signal of the double circle, but by interpolation, it is possible to shift to the negative position. For example, L
= 128, if it was R-32, everything would be possible.

Furthermore, when taking out the signal from the memory (13), the clock frequency is increased four times that of the conventional one, and the DA conversion circuit (13) is
4), or use memory (1
The signal from 3) is decimated to - and the same DA conversion as before is performed.

p' Therefore, for example, when R' = 125, processing times (12
), (13), a signal is extracted by synthesis with a shift amount of R'=125, and a signal whose time has been returned to the same level as the original signal is extracted.

In this way, the speed and pitch of the audio signal can be changed arbitrarily. In this case, the amount of calculation for the Fourier transform and the inverse transform is the same as in the conventional method, and by performing interpolation after the inverse transform, the unit of change in the shift amount becomes finer.

Further, by performing phase adjustment in advance according to the interpolated signal, good control without deterioration of sound quality can be performed.

Although a larger memory capacity is required for the interpolation during synthesis, the amount of hardware involved is not a problem compared to the arithmetic unit.

Effects of the Invention According to the present invention, it has become possible to perform fine speech control in units of change without increasing the amount of calculations.

[Brief explanation of the drawing]

FIG. 1 is a diagram for explaining a conventional device, FIGS. 2 to 5 are diagrams for explaining a speech control device previously proposed by the inventor of the present application, and FIG. 6 is a configuration of an example of the present invention. 7 are diagrams for explaining the same. (1) is an input terminal, (2) is an AD conversion circuit, (3), (
4), (12), (13) are Bazza memory, (5),
Zeng, (9) and (11) are multipliers, (6) is a Fourier transform circuit, (8) is a processing circuit, α0) is an inverse Fourier transform circuit, (14) is a DA conversion circuit, and (15) is an output terminal. ,(3
1) is an interpolation circuit. Teusane City Seishuku 2 May 10, 1980 Director-General of the Patent Office Kazuo Wakasugi Tono Meji 1, Indication of the Case 1982 Patent Application No. 250781 2°Q I
JJ O)'r511; -751
3. Relationship with the case of the person making the amendment Patent applicant address: 6-7-35, Kitashina-yo, Tokyo Parts-Yo-ku Name (2
'1B) Representative Director of Sony Corporation Norio Ohga 4, Agent Address: 1-8-1 Nishi-Shinjuku, Shinjuku-ku, Tokyo 03-
343-5821&0 (Shintaku Building) 6. Number of inventions increased due to sleeve correction (1) The scope of claims is amended as shown in the attached sheet. (2) In the specification, page 3, line 2 to page 4, line 2 “Figure 2.
... will be supplied. ” is corrected as follows. 1- In Figure 2, an audio signal that has been converted into an electrical signal using a microphone or the like and passed through a low-pass filter with a cutoff frequency of 3.2kllz is input to the input terminal (1).
supplied to This input audio signal is 6', 4kllz
An AD converter (2) of 12 bits per word driven by a conversion clock (periodically 158 μs) sequentially converts each word into 12 bits of digital data at the rate of this clock pulse. AD converter (2) is 6.4kll
1i driven by the clock of z! It is connected to a 256-word shift register (3) consisting of 12 bits,
Every time one pulse of the driving clock is supplied to the shift register (3), the shift register (3) outputs 1 if! ,
In FIG. 2, the output data of the AD converter (2) is shifted to the right (the words "left" and "right" will be used to mean left and right in FIG. 2). 1 word, from the left of shift register (3), shift register (3)
) There are people in In other words, the shift register (3) has an AD converter (2).
A series of 256 words of digital data generated by the AD converter (2) is input, and each time the AD converter (2) generates one word of digital data, the shift register (3)
The word is shifted to the right and its contents are updated. Here, before explaining the specific flow of signals below (4) in FIG. 2, general matters regarding short-time Fourier analysis will be described. For example, if we consider the audio signal "Aiueo", the time when the sound "A" is being made and the time when the sound "1i" is being made differ depending on the mouth of the person making the sound. The shape of the vocal tract is different. In other words, the audio signal "Aiueo" is a signal emitted from a physical entity whose characteristics change over time, and cannot be regarded as a stationary signal. In this way, the characteristics of the physical entity that emits audio signals, music signals, etc. change over time.
In general, it cannot be regarded as a stationary signal, and it is impossible to directly apply Fourier spectrum analysis to stationary signals. However, let's talk about which examples of "aiueo" are "a", "1-i", "u", and "1".
Yeah.” The shape of the human mouth and vocal tract remains fairly constant during the time each "o" sound is uttered, and if the signal is limited to that time, it can be regarded as a steady signal. Therefore, if we limit the region to be Fourier transformed to a time interval that can be considered stationary, perform Fourier transform, and use the Fourier spectrum obtained by updating that interval one after another, it is possible to Fourier analysis becomes possible for audio signals and music signals that are stationary in the interval. This kind of Fourier analysis is called short-time Fourier analysis. Let's explain further using mathematical formulas. Let the human signal x (tl be the data string obtained by sampling (X(→)(m=
0.1.2. ), the above-mentioned matters are
Subsequence of data that can be considered stationary (x (m+SR))
m=0.1. −...; S =0.1. ...= (R,
For a variable m of a certain integer constant M, a finite subsequence (x
(m+SR)) m=0+L..., window coefficient (h(-m
)) After multiplying by (m=0.1...,M-1),
Discrete Fourier transform of °ζ is performed on the variable m, and the short-time Fourier spectrum X (SR, k) (S=0.
1. ...+M-1; = 0.1, 2. ..., M
-1). 2π As is clear from Figure 8, when analyzed, R is the length of the interval F
iWk, and has the following restrictions. From formula (A), if we set 2π m+5R=f, then 2π − (B) Window coefficient (h (mth (rn-0, 1, 2,”, M-1
) is expanded from 1 to 10 flashes with respect to m and set to 9π - (C).In other words, X(SR,k) becomes the first
For the th variable SR, the data string is convolved with the data string +(m)), X(S, k)(S=0.1.
2. The old ) is resampled every R-1 data, and the digital signal impulse response (
h(m)) It can be interpreted that the output input to the linear digital system is resampled every R-1 data. Therefore, the update 1iRXl of the interval to be analyzed must be in the band for the first variable m, as shown by the sampling theorem. In the band of (X (m, k) ) (m=0.1.2..."), it depends on (m = 01112...), but its upper limit is shown in the figure. Since it is suppressed by the low-pass characteristic of a linear digital system with an impulse response (h(m)) at , Lt2.-”
) band + 1J) --(D) That is, R must be 1 (E). As an example, M=256. If (h (m) ) is the Hamming window coefficient, the window coefficient b (rn) = 0.5
4-0.46 cos(2πm/255) (m=
0.11・−, 255), then (h (m
)) In the low-pass band of (m=0.1, 2. -, 255), it is attenuated to about 42 dB.In other words, from the relationship in the above equation, R must be R≦□-64 ■ . In FIG. 2, the short-time Fourier transform described above is performed in (41, (51, (61, (71).M =
256, Hamming window coefficient h(rn
)-0,54-0,46Xcos (2πm/255)
(m-0,1,2°...+255), R=
It is set at 64. As is clear from the example above, R=6
4 satisfies formula (E). The details will be explained below. A shift register (12 bits per word, 256 words)
The content of 3) is that the drive clock of the AD converter (2) is set to 64
One pulse of the divided clock (i.e. 64x (AD
Drive clock period of conversion (2), approximately 158μ5ec) (
Similarly, each word is launched into a shift register (4) consisting of 256 words, 12 bits per second). The latched 256 pieces of data are transferred to the 8M1lz (period 125 n 5ec) gate C17 which is supplied to the shift register (4).
At the same timing, the data is shifted one word to the right and fed into one input terminal of a multiplier (5) having two input terminals of 12 bits and one output terminal of 23 bits. On the other hand, at this same clock timing, R
Hamming window coefficient h (m) = 0. stored in OM.
54-0.46cos (2πm/255) (r
n = 0.1, 2゜..., 255) are carried in the order of m-(L1+2+...), word by word, and the product of these two human forces is calculated by the multiplier (5). As the output,
The human input data is set to the output terminal of the multiplier (5) after 100 n sec from ζ. At the timing of the timing clock that sends human data (every 125 nsec) F F T (Fast Fourier T
transform) is sent to the converter (6). FF
When the data of 23 bits per word sent in this way becomes 256 words, the T converter (6) converts the 23 bits per word,
ζ, FFT is performed on the data of 256 words, and the real part is
Generates 256 words of complex data in which both the imaginary part consists of 16 bits. Now, the input data of 256 words to the FFT converter (6) is (y(m)) (m=0.1...+255
) output data (Y (k) (k=0.1.2...
..., 255), then from the definition of FFT, 2π - (F) On the other hand, this human data (y (In) ) (rn=o
From equation (A), the short-time Fourier spectrum of lil..., 255) is 2π but 2, and (Y (10) (k=0.1.2...
..., 255) and (X (64S, k )) (k=0
．． 1.2. ..., 255) has the following relationship: 2π (k=0.1, 2..., 255) - (H). Therefore, output 2 of FFT converter (6)
π A short-time Fourier spectrum of the human power data X(m) will be obtained. This is done by the multiplier (7). In other words, the 256 words of complex data generated by the FFT converter (6), consisting of 16 bits for both the real and imaginary parts, are as follows:
One of the multipliers (7) which has two complex data input terminals each consisting of 16 bits for both the real and imaginary parts and one output terminal consisting of 16 bits for both the real and imaginary parts at the timing of a clock with a period of 125 n see. is sent to the input terminal of On the other hand, at the timing of this same clock, the above-mentioned coefficients, 2π 2,..., 255), which have been pre-windowed, are transferred - word by word to the other input terminal of the multiplier (7). In addition, the product of these two human forces is the output of the multiplier (7),
The input data is set to the output terminal of the multiplier (7) after 100 n sec from °ζ. This output result is sent word by word to the spectrum modification circuit (8), 256 words in total, at the timing of the clock that sends the input data to the multiplier (7). (3) Same, page 4, line 3, page 5, line 20, page 17, line 7-
In each of the 8th line, the words "processing circuit" are corrected to "Spec I/RAM transformation circuit." (4) Same, 4th shell, line 7 to page 5, line 6 “This processing circuit...
...is taken out. ” is corrected as follows. 1゛The complex data of 256 words, each word consisting of 16 bits for both the real and imaginary parts, transformed by the spectrum transformation path (8) is as follows: (9), (11), (12), (1
3). The signal is converted into a time domain signal in (14). Before specifically explaining the flow of (9) to (14), (9)
) to (14), the general relationships will be described below. As mentioned earlier, the transformed short-time Fourier spectrum X (Slj', k) (S=0.1.2.
...; k = 0.1.2. ..., Tol) is,
Short-time Fourier spectrum X (S, k) (S=0
．． 1, 2. . . .;k'''Otl+2, . Therefore, the transformed short-time Fourier spectrum X (SR', k
) (S=0.1, 2....1k=(Ll, 2..
..., To create a time domain signal from
X (SR', k) (S=0.1.2...
;to=0.1,2. ..., Tol) is interpolated, and X
(S, k) (S=0.1.2...; k =0
．． 1, 2. ..., make G1) and X (S, k)
(S=0.1.2..., k=0.1.2.
.... 1) may be subjected to inverse discrete Fourier transform. That is,
Regarding the first variable of X (SR', k), ``ζ,
For each, add R'-1 0 between adjacent data ^ data
．． 1. -..., G1) is passed through a low-pass filter with C as an impulse response to produce -X(S, k). From the definition of the expression % expression % ′zo(β, k), m= −■ After this, let X(S, k) be the second variable, and with respect to k,
Perform inverse discrete Fourier transform to obtain output signal (y(S))(S
=0.1.2. ...) is obtained. If we also write this as a formula,
It will look like this: 2π - (I) R' = R and when the spectrum is not manipulated, the input signal must become the output signal as it is. For that purpose, from the above formula, y (S) = x (S) By the way, since , we set A = S - pM (p: variable) = (J) Therefore, (h (m)) and (f ( m) ) is all S
It is necessary that m=-■-(K). Now, from equation (I), if we write...) (f (m)) is m = 0. L2+”
” Since t is not 0 only at M-1, f (S-mR') ・x (mR', S) is S=
mR' + mR' + 1+ ..., mR'
+M −1(n = S−mR' + n = Ot1.・
・"+M-1) is not 0. Therefore, if R' is chosen to divide M, r-R'=M (r: positive integer constant), (m-1)R '+y1≦S≦mR'+M-1 (m=0
．． l, 2. ...), and with a finite number of additions (-y(S)) (S=0.1.2
．． ...) are found sequentially. Also, to use FFT when calculating x (mR+s),
There is a relationship of equation (H) between FFT-transformed data and short-time Fourier spectrum data 2π 0.11..., M-1i S-0,1,2,...
...; R', an integer constant), and then perform FFT. This will be explained in detail in FIG. In the following explanation, R
'=64. The real part, transformed by the spectrum transformation circuit (8),
Short-time Fourier spectrum of 256 words with both imaginary parts consisting of 16 bits X (64S, k) (k=0.1.2
゜..., 255) is a period of 125 n sec
At the clock timing of k=o, 1.2.・・・・・・
The power u (9
). On the other hand, the same clock timing is prepared in advance. The above coefficients 9π...+255) are sent one word at a time in the order of -0, 1, 2,... to the other input terminal of the multiplier (9), and the product of these two inputs is , as the output of the multiplier (9), the input data is sent to the multiplier (9) and then 1
After 00 n seconds, it is set to the output terminal of the multiplier (9). This output result is a total of 256 words, word by word, at the timing of the clock that sends input data to the multiplier (9).
Words are sent to the inverse FFT converter (lffl).The inverse FFT converter outputs 256 words, both the real and imaginary parts of which are 16 bits.
An inverse FFT is performed on this data to generate time domain data of 256 words each consisting of 16 bits. This 256-word data consisting of 16 bits per word has a period of 1
At a clock timing of 25 n sec, it is fed into one input terminal of a multiplier (11) having two input terminals each consisting of 16 bits and one output terminal consisting of 16 bits. On the other hand, at the timing of this same clock, the multiplier (11
), the above-mentioned relational expression (K) prepared in advance in the ROM is m=LL2+...
The product of these two inputs is input to the multiplier (11) 100 n seconds after the input data is set.
is set to the output terminal of This output result is 256 words in total, one word at a time at the timing of the clock that sends input data to the multiplier (11).
It is sent to the shift register (12). The shift register (12) consists of 256 words of 16 bits per word, and sends out the multiplication results of the multiplier (11). They are driven by the same clock with a period of 125 n sec, and each time the multiplication result is carried over from the multiplier (11) by one word, it is shifted to the right by one word. In this way, when the multiplication result of the multiplier (11) of 256 words is entered into the shift register (12), the shift register (12) becomes in a shift prohibited state and the 256 words of the shift register (12) become 1 Each of the shift registers (13), each consisting of 16 bits and 256 words, is added for each corresponding word, and the addition result is placed into the corresponding word of each shift register (13). This shift register (13) is supplied with a 6.4kHz clock that drives the AD converter (2), and when the above-mentioned addition is completed, this 6.4kHz clock is shifted every pulse. The register (13) is shifted one word to the right and l data is sent to the 16-bit AD converter (14). On the other hand, due to this shift, data having a value of 0 is transferred to the shift register (13) from the left to 1.
It can be put into words. In this way, the shift register (13) performs shifting R'=64 times and sends 64 output data to the DA converter. The 16-bit DA converter (14) receives li! which is sent at a clock timing of 6.4kllz. The 16-bit data is sequentially converted to an analog voltage value, and the output terminal (
15). ” (5) Same, page 6, line 1, “The phase...signal” is 1. "This signal," he corrected. (6) Same page, lines 11 and 15, and Susha, page 7, line 3. (7) Similarly, on the 7th page, IO line, after "for", add "multiply by window coefficient h(m)". (8) Same page, line 13, after converting by 1 degree, 1-spectrum X2 (SR, ω) is added after ". (9) Same, same page, line 16 is corrected. (lO) Same, page 8, lines 7-14 1h(m)=1z...can be done. ” is corrected as follows. rh (m) = 1 f (m) = 0.5-0.5cos (2πm/
(N 1) >m=0,...N-1 When the low frequency component is larger, h (m) = 0.54-0.46 cos (2yc
The sound quality can be improved by setting m/(N-1))m=0,...N-1 f(m)=R'/Σh1. Note that h (m)=O corresponds to doing nothing with the multiplier (5). "(11) Same, page 9, lines 6 to 9, 1 value of 1 phase... phase" should be corrected as follows. [In this case, if one value of the phase is P (SR, ωk), P (SR, ωk) takes a value of -π≦P' (SR, ωk) < π, and the phase becomes discontinuous. Part exists. Therefore, the phase after removing this discontinuity is added.'' (12) After ``Further'' on page 10, line 10, ``To determine the discontinuous portion.'' is added. (13) In the same article, page 13, line 11 to page 14, line 1, line 1 - ``Speech control device operated manually'' is corrected as follows. 1- Means for extracting a human-generated audio signal by multiplying it at each arbitrary time interval by a window function having a length and coefficient longer than the time limited by the time interval, and Fourier transform for each extracted signal. means for converting the time axis into the frequency axis by using the method; means for adjusting the phase of the converted signal; means for inversely converting the phase-adjusted signal from the frequency axis to the time axis; means for interpolating the inversely transformed signal at a predetermined magnification, multiplying the interpolated signal by a window function having a time length and a coefficient defined by the window function, and sequentially synthesizing the signals at each arbitrary time interval; "A speech control device comprising means for arbitrarily expanding or contracting the time axis and outputting it." (16) Same, page 17, line 1, ``Figure 5'' is corrected to ``Figure 5, Figure 8, Figure 9''. (17) Figures 8 and 9 will be added to the drawings as shown in the attached sheet. What is claimed is: a means for multiplying and extracting an input audio signal for each arbitrary time interval by a window function having a length and a coefficient longer than the time limited by the time interval, and for each extracted signal. means for converting the time axis into the frequency axis by Fourier transform; means for adjusting the phase of this converted signal; and means for inversely converting the frequency axis into the time axis by performing inverse Fourier transform on the phase-adjusted signal. , a means for interpolating this inversely transformed signal at a predetermined magnification, and a means for multiplying this interpolated signal by a window function having a time length and a coefficient defined by the window function 2. A speech control device comprising: means for sequentially synthesizing each segment, and arbitrarily expanding/contracting the time axis and outputting the output.

Claims (1)

[Claims] A means for extracting an input audio signal by an arbitrary time length at an arbitrary time interval, a means for performing a Fourier transform on each extracted frame and converting the °ζ time axis into a frequency axis, means for adjusting the phase of the signal; means for performing inverse Fourier transform on the phase-adjusted signal to inversely transform the ′ζ frequency axis into the time axis; and means for interpolating the inversely transformed signal at a predetermined magnification; A speech control device comprising means for sequentially synthesizing the interpolated signals at desired time intervals and arbitrarily expanding or contracting the time axis for output.