JPH05265486A - Speech analyzing and synthesizing method - Google Patents

Abstract

PURPOSE:To improve the quantity of a speech by predicting the terminal phase of a block according to pitch information in each block of a speech signal transmitted from analysis side and the block initial phase, and correcting the phase by using Gaussian noises. CONSTITUTION:An analysis part 10 takes the input speech out in blocks of a specific number of samples and extracts pitch information from the speech signal in the blocks. The speech signal in the blocks is converted into a frequency axis and power information and voiced-sound/voiceless-sound decision information are found in each of bands divided according to the pitch information. A synthesis part 20 synthesizes speeches by a voiced-sound synthesis part 21 and a voiceless-sound synthesis part 27 according to the pitch information sent from the analysis part 10, etc.; and they are added by an addition part 28 and a synthesized speech signal is outputted. The phase prediction part 22 of the voiced-sound synthesis part 21 predicts the block terminal phase according to the pitch information and block initial phase and a phase correction part 24 corrects the terminal phase by using the Gaussian noises having variance corresponding to the respective bands.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech analysis / synthesis system applied to a speech signal analysis / synthesis coding apparatus.

[0002]

2. Description of the Related Art Human hearing is a kind of spectrum analyzer, and has the property that it sounds as the same sound if the power spectra are equal. A method of obtaining a synthetic sound by utilizing this property is a voice analysis / synthesis method.

In order to obtain the above-mentioned synthesized sound, the analyzing side analyzes the input voice signal, extracts or detects pitch information, voiced / unvoiced sound discrimination information, amplitude information, etc., and transmits them to the synthesizing side. A voice is artificially created based on such information. In particular, the synthesizing side can be classified into a recording editing method, a parameter editing method, a rule synthesizing method, etc., depending on the synthesizing method.

In the above-mentioned recording / editing method, the voice generated by a person is accumulated (recorded) in units of words, phrases, etc. in advance, read out as needed, and connected (edited).
It synthesizes voice.

The parameter editing method uses words, phrases, and the like as a unit as in the case of the recording editing method. However, the voice uttered by a person is analyzed in advance based on the voice generation model to form the parameter time series. In this method, a voice synthesizing device is driven by using a parameter time series that is stored as described above and is connected as needed to synthesize a voice.

The rule synthesizing method is a technique for continuously converting a sequence represented by discrete symbols such as characters and phonetic symbols. In the process of conversion, universal and artificial properties of speech production are applied as synthesis rules.

In each of the above synthesis methods, the vocal tract characteristic is simulated in some form, and a synthesized sound is obtained by using a signal having substantially the same spectrum as the sound source wave.

[0008]

By the way, in the above speech analysis / synthesis method, it is necessary to match the phase on the synthesis side with the phase on the analysis side. In this case, when the phase information is obtained on the combining side, linear prediction by angular frequency and correction by white noise may be used. However, control of noise (error) based on the true value of the phase and prediction is impossible with the white noise.

Further, since the level of white noise is changed and used as a correction term at a ratio of unvoiced sound in the entire band,
When a block containing a lot of voiced sound is continuous, no correction is made only by prediction. As a result, when a strong vowel lasts for a long time, errors are accumulated and sound quality deteriorates.

Therefore, it is an object of the speech analysis / synthesis method according to the present invention to provide a speech analysis / synthesis method for improving the sound quality by using noise whose magnitude and variance can be controlled for correction of prediction. ..

[0011]

A speech analysis / synthesis method according to the present invention comprises a step of dividing an input speech signal into blocks to obtain pitch information within the block, and a frequency of the speech signal for each block. Axis to obtain frequency-axis data, dividing the frequency-axis data into a plurality of bands based on the pitch information, power information for each divided band, and whether voiced or unvoiced And the step of transmitting the above-mentioned pitch information obtained by these steps, the power information for each band, and the discrimination information of voiced sound or unvoiced sound, and for each block obtained by the transmission. Predicting the block end phase based on the pitch information and the block initial phase, and modifying the predicted block end phase using noise having a variance according to each band To solve the above problems and a that step.

The speech analysis and synthesis method according to the present invention is
The above problem is solved by the characteristic that the noise is Gaussian noise.

[0013]

According to the speech analysis and synthesis method of the present invention, the frequency-axis data obtained by converting the speech signal of each block into the frequency axis is divided into a plurality of bands based on the pitch information obtained from the speech signal of each block. The analysis side obtains the power information and discrimination information of voiced sound or unvoiced sound for each and transmits it, and the synthesis side determines the block end phase based on the pitch information and the block initial phase for each block obtained by the transmission. It is possible to control the error between the predicted phase value and the true value by predicting and correcting the predicted terminal phase using Gaussian noise having a variance according to each band.

[0014]

DESCRIPTION OF THE PREFERRED EMBODIMENTS A specific example in which the speech analysis / synthesis method according to the present invention is applied to a speech signal analysis / synthesis coding apparatus (so-called vocoder) will be described below with reference to the drawings.
This analysis-synthesis coding apparatus performs modeling such that a voiced sound section (Voiced) section and an unvoiced sound section (Unvoiced) section exist in the frequency domain at the same time (in the same block or frame).

FIG. 1 is a diagram showing an overall schematic configuration of an embodiment in which the present invention is applied to the above-described speech signal analysis / synthesis coding apparatus. In FIG. 1, an embodiment of the voice analysis / synthesis method according to the present invention comprises an analysis unit 10 for analyzing pitch information and the like from an input voice signal, and various information (pitch information transmitted by the transmission unit 2 from the analysis unit 10). Information etc.),
The synthesis unit 20 synthesizes voiced sound and unvoiced sound, and further synthesizes the voiced sound and unvoiced sound.

The analyzing unit 10 extracts a voice signal input from the input terminal 1 in block units of a predetermined number of samples (N samples), and an input voice signal for each block from the block extracting unit 11. From this, a pitch information extraction unit 12 for extracting pitch information, a data conversion unit 13 for obtaining data converted on the frequency axis from the input voice signal for each block from the block extraction unit 11, and this data conversion unit 13 Band division unit 14 for dividing the data on the frequency axis into a plurality of bands based on the pitch information of the pitch information extraction unit 12, and power (amplitude) information and voiced sound V for each band of the band division unit 14. Amplitude information & V / U for determining unvoiced sound UN information
The V discrimination information detection unit 15 is included.

The synthesizing unit 20 has the pitch information, V / UV, transmitted from the analyzing unit 10 by the transmitting unit 2.
Upon receiving the discrimination information and the amplitude information, the voiced sound synthesizing unit 21 synthesizes the voiced sound with the unvoiced sound synthesizing unit 27, and the synthesized voiced sound and unvoiced sound are added and synthesized by the adding unit 28 to obtain the synthesized sound signal. It is taken out from the output terminal 3.

The above information is obtained by processing the data in the block of N samples (for example, 256 samples), but the block is the frame of L samples on the time axis. Since the data advances in units, the data to be transmitted is obtained in frame units. That is, pitch information, V
/ UV discrimination information and amplitude information will be updated.

The voiced sound synthesizing section 21 predicts the frame end phase (the phase of the leading edge of the next synthesized frame) based on the pitch information and the frame initial phase supplied from the input terminal 4, and , The phase predictor 22
Based on the corrected phase information from the phase correction unit 24 and the phase correction unit 24 that corrects the prediction from the above using the correction term from the noise adding unit 23 to which the pitch information and the V / UV discrimination information are supplied. The sine wave generator 25 reads out and outputs a sine wave from a sine wave ROM (not shown), and the amplitude amplifier 26 is supplied with the amplitude information and amplifies the amplitude of the sine wave from the sine wave generator 25.

The unvoiced sound synthesizer 27 is supplied with the pitch information, V / UV discrimination information and amplitude information. For example, white noise is filtered by a bandpass filter (not shown) to synthesize an unvoiced sound waveform on the time axis. There is.

In the adding section 28, the voiced sound synthesizing section 2 is used.
1. The voiced sound and unvoiced sound signals synthesized by the unvoiced sound synthesizer 27 are added at an appropriate fixed mixing ratio. Then, the added audio signal is output from the output terminal 3 as an audio signal.

Here, the voiced sound synthesizer 2 of the synthesizer 20 is used.
In the phase prediction unit 22 within 1, the time 0 (the beginning of the frame)
The phase of m-th harmonic (initial phase of frame) at ψ _0m
Then, the phase ψ _Lm at the end of the frame is predicted as ψ _Lm = ψ _0m + m (ω _O1 + ω _L1 ) L / 2 (1). The phase φ _{m of} each band is φ _m = φ _Lm + ε _m (2) In the above equations (1) and (2), L is the frame interval, ω _O1 is the fundamental angular frequency at the leading edge (n = 0) of the composite frame, and ω _L1 is the end (n = L: The fundamental angular frequency at the composite frame tip), ε _m , indicates the prediction correction term in each band.

From the equation (1), the phase predictor 22
Calculates the phase obtained by multiplying the average angular frequency of the m-th harmonic by the time and adding the initial phase of the m-th harmonic as the predicted phase at the time L. Further, from the above formula (2), the phase φ _{m of} each band is a value obtained by adding the prediction correction term ε _m to the above-mentioned predicted phase.

The predictive correction term ε _m has a random (random) distribution among the bands, and random numbers can be used, but Gaussian noise is used in this embodiment. As shown in FIG. 2, this Gaussian noise is a noise in which the dispersion increases as the frequency becomes higher (for example, ε ₁ to ε ₁₀ ) in each band. This Gaussian noise properly approximates the error between the true value of the phase and the predicted value.

Now, assuming that the variance shown in FIG. 2 is simply proportional to m in the band, the prediction correction term ε _m is ε _m = h ₁ N (0, k _i ). .. (3) is indicated. Here, h ₁ is a constant, k _i is a fraction, and 0 is an average.

Further, when the entire band is divided into two bands of voiced sound and unvoiced sound, if there are many unvoiced sounds, the phase of each frequency component forming the voice becomes more random.
The prediction correction term ε _m can be expressed as ε _m = h ₂ n _uj N (0, k _i ) ... (4). Here, h ₂ is a constant, k _i is a fraction, 0 is an average, and n _uj is the number of unvoiced bands in block j.

In particular, when there is no disturbance in the distribution between the bands as described above when the vowel of the input voice continues for a long time, or when the vowel changes to consonant and silence, the above (3), ( Since the predictive correction term expressed by the equation (4) rather deteriorates the sound quality of the synthesized voice, if delay is allowed, the amplitude information (power) S level one frame ahead or the decrease of the voiced sound portion is checked to determine the above. Correction term ε _{m
The, ε m = h 3 max (} a, S j -S j + 1) N (0, k i) ··· (5) ε m = h 4 max (b, n vj -n v (j + 1 ₎ ) N (0, k _i ) ... (6) Here, a, b, h ₃ and h ₄ are constants.

Further, when the pitch information in the pitch information extraction unit 12 is low, the above correction term ε _m is set to ε _m = f in consideration of an increase in adverse effects due to increase in frequency bands and alignment of phases. (S _j , h _j ) N (0, k _i ) ... (7). Where f is the frequency

As described above, in the embodiment in which the present invention is applied to the speech signal analysis / synthesis coding apparatus, the noise used for the correction of the phase prediction is Gaussian to control the magnitude and variance thereof. ..

The speech analysis and synthesis method according to the present invention will be described below.
A specific example applied to an MBE (Multiband Excitation) vocoder, which is a kind of speech signal synthesis analysis coding apparatus (so-called vocoder), will be described with reference to the drawings. This MBE vocoder is DW Gr
iffin and JS Lim, "Multiband Excitation Vocode
r, "IEEE Trans. Acoustics, Speech, and Signal Process
ing, vol.36, No.8, pp.1223-1235, Aug.1988, the conventional PARCOR (PARtial au
In a to-CORrelation (partial autocorrelation) vocoder or the like, a voiced sound section and an unvoiced sound section were switched for each block or frame when modeling a voice, whereas in the MBE vocoder, at the same time (same block or frame). The model is based on the assumption that there are voiced and unvoiced intervals in the frequency domain of (in).

FIG. 3 is a block diagram showing an overall schematic configuration of an embodiment in which the present invention is applied to the MBE vocoder. In FIG. 3, an audio signal is supplied to the input terminal 101, and the input audio signal is H
It is sent to a filter 102 such as a PF (high-pass filter) to remove a so-called DC (direct current) offset and at least a low frequency component (200 Hz or less) for band limitation (for example, 200 to 3400 Hz). .. The signal obtained through the filter 102 is sent to the pitch extraction unit 103 and the windowing processing unit 104, respectively. The pitch extraction unit 103 divides the input voice signal data into blocks in units of a predetermined number N (for example, N = 256) (or cuts out with a rectangular window), and pitches the voice signals in this block. .. Such cut-out block (256 samples) is divided into L samples (eg L = 16) as shown in A of FIG.
0) frame intervals are moved in the time axis direction, and the overlap between blocks is NL samples (for example, 9 samples).
6 samples). Also, the windowing processing unit 104
In this case, a predetermined window function, for example, a Hamming window is applied to one block of N samples, and this windowed block is sequentially moved in the time axis direction at intervals of one frame of L samples.

When this windowing process is expressed by a mathematical expression, x _w (k, q) = x (q) w (kL-q) (8) In this equation (8), k is a block number,
q represents the time index (sample number) of the data, and the q-th data x (q) of the unprocessed input signal is windowed by the window function (w (kL-q)) of the kth block. It is shown that the data x _w (k, q) can be obtained by doing so. FIG. 4A in the pitch extraction unit 103
The window function w _r (r) in the case of a rectangular window is as follows: w _r (r) = 1 0 ≦ r <N (9) = 0 r <0, N ≦ r window function w _h (r) in the case of Hamming window as shown in B of FIG. 4 is a section _{104, w h (r) =} 0.54 - 0.46 cos (2πr / (N-1)) 0 ≦ r <N (10) = 0 r <0, N ≦ r. Such a window function w _r (r) or w (8) when using the _h (r) formula of the window function w (r) (= w (KL-
q)), the zero-zero interval is 0 ≦ kL−q <N, which is transformed into kL−N <q ≦ kL. Therefore, for example, in the case of the above rectangular window, the window function w _r (kL−q) =
As shown in FIG. 5, 1 becomes kL-N <q ≦ kL.
It will be when. Further, the above equations (8) to (10) are expressed by the length N
The window of (= 256) samples is shown to advance by L (= 160) samples. Below, above (9)
N points (0 ≦ r
The non-zero sample sequence of <N) is represented by x _wr (k, r) and x _wh
Let us denote it as (k, r).

In the windowing processing unit 104, as shown in FIG. 6, 179 is applied to the sample sequence x _wh (k, r) of 256 samples of one block on which the Hamming window of the above equation (10) is applied.
Two samples of 0 data are added (so-called zero padding) to make 2048 samples, and the orthogonal transformation unit 105 performs orthogonal transformation such as FFT (Fast Fourier Transform) on the time-axis data sequence of 2048 samples. Processing is performed.

In the pitch extraction unit 103, the above x _wr (k, r)
Pitch extraction is performed based on the sample sequence (1 block N samples). The pitch extraction method is known to include periodicity of time waveform, periodic frequency structure of spectrum, and autocorrelation function.
The center correlation waveform autocorrelation method is used. Regarding the center clip level in the block at this time, one clip level may be set for one block, but the peak level of the signal of each part (each sub-block) obtained by subdividing the block is detected and When there is a large difference in the peak level of each sub-block, the clip level is changed stepwise or continuously within the block. The peak period is determined based on the peak position of the autocorrelation data of the center clip waveform. At this time, a plurality of peaks are obtained from the autocorrelation data belonging to the current frame (the autocorrelation is obtained for the data of N samples of one block), and the maximum peak among the plurality of peaks is equal to or larger than a predetermined threshold. In the case of, the maximum peak position is set as the pitch period, and in other cases, the pitch is within a pitch range that satisfies a predetermined relationship with the pitch other than the current frame, for example, the pitch of the previous frame and the pitch of the previous frame. As a result, a peak in the range of ± 20% is obtained, and the pitch of the current frame is determined based on this peak position. In this pitch extraction unit 103, a relatively rough pitch search is performed by an open loop, and the extracted pitch data has a high precision (fine) pitch search unit 10.
Then, the high precision pitch search (pitch fine search) is performed by the closed loop.

High precision (fine) pitch search unit 106
Includes rough pitch data of integer (integer) values extracted by the pitch extraction unit 103 and the orthogonal transformation unit 10.
5, the data on the frequency axis subjected to FFT, for example, is supplied. In this high precision pitch search unit 106,
Centering on the above coarse pitch data value, it is ± 0.2 in increments of ±
Shake several samples at a time to drive to the optimum fine pitch data value with a decimal point (floating). As a fine search method at this time, a so-called analysis by synthesis method is used, and the pitch is selected so that the synthesized power spectrum is closest to the power spectrum of the original sound.

The fine search of the pitch will be described. First, in the MBE vocoder, the FF
Let S (j) as spectrum data on the frequency axis orthogonally transformed by T etc. be expressed as S (j) = H (j) | E (j) | 0 <j <J (11) It is assumed that the model. Here, J corresponds to πω _s = f _s / 2, and the sampling frequency f _s =
When 2πω _s is, for example, 8 kHz, it corresponds to 4 kHz.
In the equation (11), when the spectrum data S (j) on the frequency axis has a waveform as shown in A of FIG. 7, H (j)
Is the original spectrum data S (j) as shown in B of FIG.
Shows the spectral envelope of E (j)
Shows a spectrum of an excitation signal (excitation) that is periodic at an equal level as shown in FIG. 7C. That is, the FFT spectrum S (j) is the spectrum envelope H (j) and the power spectrum of the excitation signal | E (j) |
It is modeled as the product of and.

Power spectrum of the excitation signal | E (j)
Is a spectral waveform corresponding to the waveform of one band (band) for each band on the frequency axis in consideration of the periodicity (pitch structure) of the waveform on the frequency axis determined according to the pitch. It is formed by arranging it repeatedly. The waveform for one band is obtained by performing FFT by regarding the waveform obtained by adding (filling with 0) 1792 samples of 0 data to the Hamming window function of 256 samples as shown in FIG. 6 as a time axis signal. It can be formed by cutting out an impulse waveform having a certain bandwidth on the frequency axis according to the pitch.

Next, for each band divided according to the above pitch, a value (a kind of amplitude) | A that represents the above H (j) (minimizes the error for each band) | A _m
Ask for |. Here, for example, when the lower and upper points of the m-th band (band of the m-th harmonic) are a _m and b _m , respectively, the error ε _m of the m-th band is

[0039]

[Equation 1] Can be expressed as | A _m | that minimizes this error ε _m
Is

[0040]

[Equation 2] Therefore, when | A _m | in this equation (13), the error ε _m is minimized. Such an amplitude | A _m | is obtained for each band, and the obtained amplitude | A _m | is used to obtain the error ε _m for each band defined by the above equation (12). Next, the sum Σε of all the bands of such error ε _m for each band
_{Find m} . Further, such an error sum value Σε _m of all bands is obtained for some slightly different pitches, and a pitch that minimizes the error sum value Σε _m is obtained.

That is, several kinds of vertical pitches are prepared with the rough pitch obtained by the pitch extraction unit 103 as the center, for example, in steps of 0.25. The error sum value Σε _m is obtained for each of these plural kinds of slightly different pitches. In this case, when the pitch is determined, the bandwidth is determined, and from the above equation (13), the power spectrum | S (j) |
(j) | can be used to find the error ε _m in the above equation (12), and the total sum Σε _m of all the bands can be obtained. This error sum value Σε _m is obtained for each pitch, and the pitch corresponding to the minimum error sum value is determined as the optimum pitch. As described above, the high-precision pitch search unit 106 obtains the optimum fine (for example, 0.25 step) pitch, and the amplitude | A _m | corresponding to this optimum pitch.
Is determined.

In the above description of the pitch fine search, in order to simplify the explanation, all bands are voiced (Vo
Assuming the case of iced), as described above, MBE
In the vocoder, unvoiced sound (Un
Since a model in which a voiced region exists is used, it is necessary to distinguish voiced sound / unvoiced sound for each band.

The optimum pitch and amplitude | A _m | data from the high precision pitch search section 106 is sent to the voiced sound / unvoiced sound determination section 107, and the voiced sound / unvoiced sound is discriminated for each band. NSR (noise to signal ratio) is used for this determination. That is, the NSR of the m-th band is

[0044]

[Equation 3] When this NSR value is larger than a predetermined threshold value (for example, 0.3) (error is large), | S (j) | is approximated by | A _m || E (j) | in that band. Is not good (the above excitation signal | E (j) | is unsuitable as a basis)
Therefore, the band is determined to be UV (Unvoiced, unvoiced sound). In other cases, it can be determined that the approximation is performed to some extent, and the band is determined to be V (Voiced, voiced sound).

Next, the amplitude re-evaluation unit 108 supplies the frequency-axis data from the orthogonal transform unit 105 and the amplitude | A _m evaluated as the fine pitch from the high precision pitch search unit 106.
| And each voiced sound / unvoiced sound discrimination unit 107
V / UV (voiced sound / unvoiced sound) discrimination data from The amplitude re-evaluation unit 108 re-calculates the amplitude of the band determined to be unvoiced sound (UV) by the voiced sound / unvoiced sound determination unit 107. The amplitude | A _m | _UV for this UV band is

[0046]

[Equation 4] Required at.

The data from the amplitude re-evaluation unit 108 is
Data number conversion (a kind of sampling rate conversion) unit 10
Sent to 9. This data number conversion unit 109 is for making the number constant, considering that the number of divided bands on the frequency axis differs according to the pitch and the number of data (especially the number of amplitude data) differs. is there. That is, for example, if the effective band is up to 3400 kHz, this effective band is divided into 8 bands to 63 bands according to the pitch, and the amplitude | obtained for each of these bands |
The number of data A _m | (including the amplitude of the UV band | A _m | _UV ) m _{MX + 1} also changes from 8 to 63. Therefore, the data number conversion unit 109 converts the variable number m _{MX + 1} of amplitude data into a fixed number N _C (for example, 44) of data.

Here, in this embodiment, dummy data for interpolating values from the last data in the block to the first data in the block with respect to the amplitude data for one block of the effective band on the frequency axis. Is added to expand the number of data to N _F , and then the bandwidth-limited K _OS times (for example, 8
By multiplying the number of times, the amplitude data of K _OS times the number is obtained, and the number of K _OS times the number of (((
(m _{MX + 1} ) × K _OS pieces of amplitude data are linearly interpolated and expanded to a larger number of N _M pieces (eg, 2048 pieces), and this N _M pieces
The data is thinned out to obtain the above-mentioned fixed number N _C (for example, 44
Data).

The data from the data number conversion unit 109 (a fixed number N _C of the amplitude data) is sent to the vector quantization unit 110, and a predetermined number of data is put together into a vector, and vector quantization is performed. Is given. The quantized output data from the vector quantizer 110 is output to the output terminal 1
It is taken out via 11. The high-precision (fine) pitch data from the high-precision pitch search unit 106 is coded by the pitch coding unit 115, and the output terminal 11
It is taken out via 2. Further, the voiced sound / unvoiced sound (V / UV) discrimination data from the voiced sound / unvoiced sound discrimination unit 107 is taken out through the output terminal 113. The data from these output terminals 111 to 113 are transmitted as signals in a predetermined transmission format.

Each of these data is obtained by processing the data in the block of N samples (for example, 256 samples), but the block is a frame of L samples on the time axis. , The data to be transmitted is obtained in the frame unit. That is, the pitch data, the V / UV discrimination data, and the amplitude data are updated at the above frame cycle.

Next, a synthesizing side (decoding side) for synthesizing a voice signal based on the above-mentioned respective data transmitted and obtained.
The general configuration of will be described with reference to FIG. In FIG. 8, the vector-quantized amplitude data is input to the input terminal 121, the encoded pitch data is input to the input terminal 122, and the V-value is input to the input terminal 123.
/ UV discrimination data is supplied respectively. Input terminal 12
The quantized amplitude data from 1 is the inverse vector quantization unit 12
4 and is inversely quantized, is then sent to the data number inverse transform unit 125 and is inversely transformed, and the obtained amplitude data is sent to the voiced sound synthesis unit 126 and the unvoiced sound synthesis unit 127. The encoded pitch data from the input terminal 122 is the pitch decoding unit 1
It is decoded at 28 and sent to the data number inverse conversion unit 125, the voiced sound synthesis unit 126, and the unvoiced sound synthesis unit 127. The V / UV discrimination data from the input terminal 123 is sent to the voiced sound synthesis unit 126 and the unvoiced sound synthesis unit 127.

In the voiced sound synthesis unit 126, for example, cosine (cosin
e) A voiced sound waveform on the time axis is synthesized by wave synthesis, and in the unvoiced sound synthesis unit 127, for example, white noise is filtered by a bandpass filter to synthesize the unvoiced sound waveform on the time axis, and these voiced sound synthesized waveforms are combined. The unvoiced sound synthesized waveform is added and synthesized by the addition unit 129 and is taken out from the output terminal 130. In this case, the amplitude data, the pitch data, and the V / UV discrimination data are updated and given for each frame (L sample, for example, 160 samples) at the time of the analysis, but the continuity between the frames is improved (smoothed). Therefore, each value of the amplitude data and the pitch data is set as each data value at, for example, the center position in one frame, and each data value up to the center position of the next frame (one frame at the time of composition) is interpolated. Ask by. That is,
In one frame (for example, from the center of the above analysis frame to the center of the next analysis frame) at the time of synthesis, each data value at the tip sample point and the end (the tip of the next synthesis frame)
Each data value at the sample point is given, and each data value between these sample points is obtained by interpolation.

The synthesis processing in the voiced sound synthesis unit 126 will be described in detail below. The one combined frame (L sample, for example, 160 samples) on the time axis in the m-th band (band of the m-th harmonic) determined to be V (voiced sound)
When the voiced sound for a minute is V _m (n), V _m (n) = A _m (n) cos (θ _m (n)) 0 using the time index (sample number) n in this composite frame. ≦ n <L (16) The final voiced sound V (n) is synthesized by adding (ΣV _m (n)) the voiced sounds of all the bands which are determined to be V (voiced sound) of all the bands.

A _m (n) in the equation (16) is the amplitude of the m-th harmonic wave that is interpolated from the beginning to the end of the composite frame. The simplest way is to linearly interpolate the value of the m-th harmonic of the amplitude data updated in frame units.
That is, the m-th frame at the tip (n = 0) of the composite frame
The amplitude value of the harmonic is A _0m , the end of the composite frame (n =
L: the amplitude value of the m-th harmonic at the end of the next composite frame) is A _Lm , A _m (n) = (Ln) A _0m / L + nA _Lm / L ... (17) It suffices to calculate A _m (n).

Next, the phase θ _m (n) in the above equation (16) is calculated by θ _m (0) = mω _O1 n + n ² m (ω _L1 −ω ₀₁ ) / 2L + φ _{0 m} + Δω n (18) You can ask. In the equation (18), φ _0m represents the phase of the m-th harmonic (frame initial phase) at the tip (n = 0) of the composite frame, and ω ₀₁ is at the tip of the composite frame (n = 0). The fundamental angular frequency, ω _L1, represents the fundamental angular frequency at the end of the combined frame (n = L: the leading end of the next combined frame). Δω in the equation (18) is
Set a minimum Δω such that the phase φ _{Lm at} n = L is equal to θ _m (L).

On the other hand, in the embodiment of the present invention, the phase is predicted and calculated on the combining side without sending φ _0m + Δωn in the equation (18) to the combining side. That is, the phase prediction unit 2
2 is obtained by adding m (ω _O1 + ω _L1 ) L / 2 to the phase (frame initial phase) ψ _0m of the m-th harmonic at time 0 (the beginning of the frame) as shown in the above equation (1). The phase ψ _{Lm at} is predicted and calculated. Further, the phase φ _{m of} each band is shown by adding ε _m to the predicted and calculated phase ψ _Lm . This ε _m indicates the prediction correction term in each band. In the present invention, Gaussian noise is used for this prediction correction term ε _m .

Here, A of FIG. 9 shows an example of a spectrum of a voice signal. Each band of band numbers (harmonics number) m is 8, 9 and 10 is UV (unvoiced sound), and other bands. Is V (voiced sound). This V
The time axis signal of the (voiced sound) band is the voiced sound synthesis unit 1 described above.
26, and the time axis signal of the UV (unvoiced sound) band is synthesized by the unvoiced sound synthesis unit 127.

The unvoiced sound synthesizing process in the unvoiced sound synthesizing section 127 will be described below. The white noise signal waveform on the time axis from the white noise generation unit 131 is windowed by a suitable window function (for example, Hamming window) with a predetermined length (for example, 256 samples), and the STFT processing unit 132 performs STFT (short circuit). By performing the (Term Fourier Transform) process, a power spectrum of white noise on the frequency axis as shown in B of FIG. 9 is obtained. The power spectrum from the STFT processing unit 132 is converted to the band amplitude processing unit 1
33, and as shown in FIG. 9C, the above-mentioned amplitude | A _m | _UV is multiplied with respect to the band (for example, m = 8, 9, 10) set as the UV (unvoiced sound), and another V (voiced sound) is generated. ) Is set to 0. This band amplitude processing unit 1
The above amplitude data, pitch data, and V / UV discrimination data are supplied to 33. The output from the band amplitude processing unit 133 is sent to the ISTFT processing unit 134, and the phase is converted into a signal on the time axis by performing inverse STFT processing using the phase of the original white noise. The output from the ISTFT processing unit 134 is sent to the overlap addition unit 135, and the overlap and addition are repeated while performing appropriate weighting (so that the original continuous noise waveform can be restored) on the time axis, and continuous. Time-domain waveforms are synthesized.
The output signal from the overlap adder 135 is sent to the adder 129.

As described above, the signals of the voiced sound portion and the unvoiced sound portion which are synthesized in the respective synthesis units 126 and 127 and are returned to the time axis are added by the addition unit 129 at an appropriate fixed mixing ratio, The reproduced audio signal is taken out from the output terminal 130.

Therefore, in the specific example in which the speech analysis / synthesis method according to the present invention is applied to MBE, the size and variance can be controlled by making the noise used for phase prediction into Gaussian.

Regarding the configuration on the voice analysis side (encoding side) in FIG. 3 and the configuration on the voice synthesis side (decoding side) in FIG. 7, although each unit is described in hardware, a so-called DSP (digital It is also possible to realize it by a software program using a signal processor or the like.

[0062]

In the voice analysis / synthesis method according to the present invention, the data on the frequency axis obtained by converting the voice signal of each block into the frequency axis is divided based on the pitch information obtained from the voice signal of each block. The analysis side obtains and transmits power information and discrimination information of voiced or unvoiced sound for each of a plurality of bands,
On the combining side, the block end phase is predicted based on the pitch information and the block initial phase for each block obtained by transmission, and the predicted end phase is a Gaussian noise having a variance according to each band. By modifying with, it is possible to control the magnitude and variance of noise, and it is expected to improve the sound quality. Further, by utilizing the signal level of the voice and its temporal change, it is possible to prevent the accumulation of errors and prevent the deterioration of the sound quality at the vowel part or at the transition point from the vowel part to the consonant part.

[Brief description of drawings]

FIG. 1 is a functional block diagram of a specific example in which a voice analysis / synthesis method according to the present invention is applied to a so-called vocoder.

FIG. 2 is a characteristic diagram for explaining Gaussian noise used in the speech analysis / synthesis method according to the present invention.

FIG. 3 is a functional block diagram showing a schematic configuration of an analysis side (encoding side) of a speech signal synthesis analysis coding apparatus as a specific example of an apparatus to which a speech analysis synthesis method according to the present invention is applied.

FIG. 4 is a diagram for explaining a windowing process.

FIG. 5 is a diagram for explaining a relationship between windowing processing and a window function.

FIG. 6 is a diagram showing time axis data as an object of orthogonal transform (FFT) processing.

FIG. 7 is a diagram showing spectrum data on a frequency axis, a spectrum envelope (envelope), and a power spectrum of an excitation signal.

FIG. 8 is a functional block diagram showing a schematic configuration on the synthesis side (decoding side) of a speech signal synthesis analysis coding apparatus as a specific example of an apparatus to which the speech analysis synthesis method according to the present invention is applied.

FIG. 9 is a diagram for explaining unvoiced sound synthesis when synthesizing voice signals.

[Explanation of symbols]

 10-Analysis unit 20-Synthesis unit 21-Voice sound synthesis unit 22-Phase prediction unit 23-Noise addition unit 24- Phase correction unit 25: Sine wave generation unit 26: Amplitude amplification unit

─────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] April 5, 1993

[Procedure Amendment 1]

[Document name to be amended] Statement

[Name of item to be corrected] Full text

[Correction method] Change

[Correction content]

[Document name] Statement

Title of Speech Analysis and Synthesis Method

[Claims]

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech analysis / synthesis system applied to a speech signal analysis / synthesis coding apparatus.

[0002]

2. Description of the Related Art Human hearing is a kind of spectrum analyzer, and has the property that it sounds as the same sound if the power spectra are equal. A method of obtaining a synthetic sound by utilizing this property is a voice analysis / synthesis method.

In order to obtain the above-mentioned synthesized sound, the analyzing side analyzes the input voice signal, extracts or detects pitch information, voiced / unvoiced sound discrimination information, amplitude information, etc., and transmits them to the synthesizing side. A voice is artificially created based on such information. In particular, the synthesizing side can be classified into a recording editing method, a parameter editing method, a rule synthesizing method, etc., depending on the synthesizing method.

In the above-mentioned recording / editing method, the voice generated by a person is accumulated (recorded) in units of words, phrases, etc. in advance, read out as needed, and connected (edited).
It synthesizes voice.

The parameter editing method uses words, phrases, and the like as a unit as in the case of the recording editing method. However, the voice uttered by a person is analyzed in advance based on the voice generation model to form the parameter time series. In this method, a voice synthesizing device is driven by using a parameter time series that is stored as described above and is connected as needed to synthesize a voice.

The rule synthesizing method is a technique for continuously converting a sequence represented by discrete symbols such as characters and phonetic symbols. In the process of conversion, universal and artificial properties of speech production are applied as synthesis rules.

In each of the above synthesis methods, the vocal tract characteristic is simulated in some form, and a synthesized sound is obtained by using a signal having substantially the same spectrum as the sound source wave.

[0008]

By the way, in the above speech analysis / synthesis method, it is necessary to match the phase on the synthesis side with the phase on the analysis side. In this case, when the phase information is obtained on the combining side, linear prediction by angular frequency and correction by white noise may be used. However, control of noise (error) based on the true value of the phase and prediction is impossible with the white noise.

Further, since the level of white noise is changed and used as a correction term at a ratio of unvoiced sound in the entire band,
When a block containing a lot of voiced sound is continuous, no correction is made only by prediction. As a result, when a strong vowel lasts for a long time, errors are accumulated and sound quality deteriorates.

[0010] Thus, a method vocoding according to the present invention has an object to provide a speech analysis-synthesis method for realizing the sound quality by using the noise can control the size and dispersion corrected prediction ..

[0011]

A speech analysis / synthesis method according to the present invention comprises a step of dividing an input speech signal into blocks to obtain pitch information within the block, and a frequency of the speech signal for each block. Axis to obtain frequency-axis data, dividing the frequency-axis data into a plurality of bands based on the pitch information, power information for each divided band, and whether voiced or unvoiced And the step of transmitting the above-mentioned pitch information obtained by these steps, the power information for each band, and the discrimination information of voiced sound or unvoiced sound, and for each block obtained by the transmission. Predicting the block end phase based on the pitch information and the block initial phase, and modifying the predicted block end phase using noise having a variance according to each band To solve the above problems and a that step.

The speech analysis and synthesis method according to the present invention is
The above problem is solved by the characteristic that the noise is Gaussian noise.

[0013]

According to the speech analysis and synthesis method of the present invention, the frequency-axis data obtained by converting the speech signal of each block into the frequency axis is divided into a plurality of bands based on the pitch information obtained from the speech signal of each block. The analysis side obtains the power information and discrimination information of voiced sound or unvoiced sound for each and transmits it, and the synthesis side determines the block end phase based on the pitch information and the block initial phase for each block obtained by the transmission. It is possible to control the error between the predicted phase value and the true value by predicting and correcting the predicted terminal phase using Gaussian noise having a variance according to each band.

[0014]

DESCRIPTION OF THE PREFERRED EMBODIMENTS A specific example in which the speech analysis / synthesis method according to the present invention is applied to a speech signal analysis / synthesis coding apparatus (so-called vocoder) will be described below with reference to the drawings.
This analysis-synthesis coding apparatus performs modeling such that a voiced sound section (Voiced) section and an unvoiced sound section (Unvoiced) section exist in the frequency domain at the same time (in the same block or frame).

FIG. 1 is a diagram showing an overall schematic configuration of an embodiment in which the present invention is applied to the above-described speech signal analysis / synthesis coding apparatus. In FIG. 1, an embodiment of the voice analysis / synthesis method according to the present invention comprises an analysis unit 10 for analyzing pitch information and the like from an input voice signal, and various information (pitch information transmitted by the transmission unit 2 from the analysis unit 10). Voiced sound based on information etc.)
And a unvoiced sound, and a synthesizer 20 for synthesizing the voiced sound and the unvoiced sound.

The analyzing unit 10 extracts a voice signal input from the input terminal 1 in block units of a predetermined number of samples (N samples), and an input voice signal for each block from the block extracting unit 11. From this, a pitch information extraction unit 12 for extracting pitch information, a data conversion unit 13 for obtaining data converted on the frequency axis from the input voice signal for each block from the block extraction unit 11, and this data conversion unit 13 Band division unit 14 for dividing the data on the frequency axis into a plurality of bands based on the pitch information of the pitch information extraction unit 12, and power (amplitude) information and voiced sound V for each band of the band division unit 14. Amplitude information & V / U for determining unvoiced sound UN information
The V discrimination information detection unit 15 is included.

The synthesizing unit 20 has the pitch information, V / UV, transmitted from the analyzing unit 10 by the transmitting unit 2.
Upon receiving the discrimination information and the amplitude information, the voiced sound synthesizing unit 21 synthesizes the voiced sound with the unvoiced sound synthesizing unit 27, and the synthesized voiced sound and unvoiced sound are added and synthesized by the adding unit 28 to obtain the synthesized sound signal. It is taken out from the output terminal 3.

The above information is obtained by processing the data in the block of N samples (for example, 256 samples), but the block is the frame of L samples on the time axis. Since the data advances in units, the data to be transmitted is obtained in frame units. That is, pitch information, V
/ UV discrimination information and amplitude information will be updated.

The voiced sound synthesizing section 21 predicts the frame end phase (the phase of the leading edge of the next synthesized frame) based on the pitch information and the frame initial phase supplied from the input terminal 4, and , The phase predictor 22
Based on the corrected phase information from the phase correction unit 24 and the phase correction unit 24 that corrects the prediction from the above using the correction term from the noise adding unit 23 to which the pitch information and the V / UV discrimination information are supplied. The sine wave generator 25 reads out and outputs a sine wave from a sine wave ROM (not shown), and the amplitude amplifier 26 is supplied with the amplitude information and amplifies the amplitude of the sine wave from the sine wave generator 25.

The unvoiced sound synthesizer 27 is supplied with the pitch information, V / UV discrimination information and amplitude information. For example, white noise is filtered by a bandpass filter (not shown) to synthesize an unvoiced sound waveform on the time axis. There is.

In the adding section 28, the voiced sound synthesizing section 2 is used.
1. The voiced sound and unvoiced sound signals synthesized by the unvoiced sound synthesizer 27 are added at an appropriate fixed mixing ratio. Then, the added audio signal is output from the output terminal 3 as an audio signal.

Here, the voiced sound synthesizer 2 of the synthesizer 20 is used.
In the phase prediction unit 22 within 1, the time 0 (the beginning of the frame)
The phase of m-th harmonic (initial phase of frame) at ψ _0m
Then, the phase ψ _Lm at the end of the frame is predicted as ψ _Lm = ψ _0m + m (ω _O1 + ω _L1 ) L / 2 (1). The phase φ _{m of} each band is φ _m = φ _Lm + ε _m (2) (1), (2) wherein L is the frame interval, Omegao1 the fundamental angular frequency at the tip of the composite frame (n = 0), omega _L1 is the composite frame termination (n = L: The following synthetic The fundamental angular frequency at the frame tip), ε _m , indicates the prediction correction term for each band.

From the equation (1), the phase predictor 22
Calculates the phase obtained by multiplying the average angular frequency of the m-th harmonic by the time and adding the initial phase of the m-th harmonic as the predicted phase at the time L. Further, from the above formula (2), the phase φ _{m of} each band is a value obtained by adding the prediction correction term ε _m to the above-mentioned predicted phase.

The predictive correction term ε _m has a random (random) distribution among the bands, and random numbers can be used, but Gaussian noise is used in this embodiment. As shown in FIG. 2, this Gaussian noise is a noise in which the dispersion increases as the frequency becomes higher (for example, ε ₁ to ε ₁₀ ) in each band. This Gaussian noise properly approximates the error between the true value of the phase and the predicted value.

[0025] Here, now, if that dependency as shown in FIG. 2 is proportional to m for simplicity band, the predicted correction terms epsilon _m is, εm = h1 N (0, ki) ··· ( 3) is indicated. Here, h ₁ is a constant, k _i is a fraction, and 0 is an average.

Further, when the entire band is divided into two bands of voiced sound and unvoiced sound, if there are many unvoiced sounds, the phase of each frequency component forming the voice becomes more random.
The prediction correction term ε _m can be expressed as ε _m = h ₂ n _uj N (0, k _i ) ... (4). Here, h ₂ is a constant, k _i is a fraction, 0 is an average, and n _uj is the number of unvoiced bands in block j.

In particular, when there is no disturbance in the distribution between the bands as described above when the vowel of the input voice continues for a long time, or when the vowel changes to consonant and silence, the above (3), ( Since the predictive correction term expressed by the equation (4) rather deteriorates the sound quality of the synthesized voice, if delay is allowed, the amplitude information (power) S level one frame ahead or the decrease of the voiced sound portion is checked to determine the above. Correction term ε _{m
The, ε m = h 3 max (} a, S j -S j + 1) N (0, k i) ··· (5) ε m = h 4 max (b, n vj -n v (j + 1 ₎ ) N (0, k _i ) ... (6) Here, a, b, h ₃ and h ₄ are constants.

Further, when the pitch information in the pitch information extraction unit 12 is low, the above correction term ε _m is set to ε _m = f in consideration of an increase in adverse effects due to increase in frequency bands and alignment of phases. (S _j , h _j ) N (0, k _i ) ... (7). Here, f is a frequency .

As described above, in the embodiment in which the present invention is applied to the speech signal analysis / synthesis coding apparatus, the noise used for the correction of the phase prediction is Gaussian to control the magnitude and variance thereof. ..

The speech analysis and synthesis method according to the present invention will be described below.
A specific example applied to an MBE (Multiband Excitation) vocoder, which is a kind of speech signal synthesis analysis coding apparatus (so-called vocoder), will be described with reference to the drawings. This MBE vocoder is DW Gr
iffin and JS Lim, "Multiband Excitation Vocode
r, "IEEE Trans. Acoustics, Speech, and Signal Process
ing, vol.36, No.8, pp.1223-1235, Aug.1988, the conventional PARCOR (PARtial au
In a to-CORrelation (partial autocorrelation) vocoder or the like, a voiced sound section and an unvoiced sound section were switched for each block or frame when modeling a voice, whereas in the MBE vocoder, at the same time (same block or frame). The model is based on the assumption that there are voiced and unvoiced intervals in the frequency domain of (in).

FIG. 3 is a block diagram showing an overall schematic configuration of an embodiment in which the present invention is applied to the MBE vocoder. In FIG. 3, an audio signal is supplied to the input terminal 101, and the input audio signal is H
It is sent to a filter 102 such as a PF (high-pass filter) to remove a so-called DC (direct current) offset and at least a low frequency component (200 Hz or less) for band limitation (for example, 200 to 3400 Hz). .. The signal obtained through the filter 102 is sent to the pitch extraction unit 103 and the windowing processing unit 104, respectively. The pitch extraction unit 103 divides the input voice signal data into blocks in units of a predetermined number N (for example, N = 256) (or cuts out with a rectangular window), and pitches the voice signals in this block. .. Such cut-out block (256 samples) is divided into L samples (eg L = 16) as shown in A of FIG.
0) frame intervals are moved in the time axis direction, and the overlap between blocks is NL samples (for example, 9 samples).
6 samples). Also, the windowing processing unit 104
In this case, a predetermined window function, for example, a Hamming window is applied to one block N samples, and this windowed block is sequentially moved in the time axis direction at intervals of one frame L samples.

When this windowing process is expressed by a mathematical expression, x _w (k, q) = x (q) w (kL-q) (8) In this equation (8), k is a block number,
q represents the time index (sample number) of the data, and the q-th data x (q) of the unprocessed input signal is windowed by the window function (w (kL-q)) of the kth block. It is shown that the data x _w (k, q) can be obtained by doing so. FIG. 4A in the pitch extraction unit 103
The window function w _r (r) in the case of a rectangular window is as follows: w _r (r) = 1 0 ≦ r <N (9) = 0 r <0, N ≦ r window function w _h (r) in the case of Hamming window as shown in B of FIG. 4 is a section _{104, w h (r) =} 0.54 - 0.46 cos (2πr / (N-1)) 0 ≦ r <N (10) = 0 r <0, N ≦ r. Such a window function w _r (r) or w (8) when using the _h (r) formula of the window function w (r) (= w (KL-
q)), the zero-zero interval is 0 ≦ kL−q <N, which is transformed into kL−N <q ≦ kL. Therefore, for example, in the case of the above rectangular window, the window function w _r (kL−q) =
As shown in FIG. 5, 1 becomes kL-N <q ≦ kL.
It will be when. Further, the above equations (8) to (10) are expressed by the length N
The window of (= 256) samples is shown to advance by L (= 160) samples. Below, above (9)
N points (0 ≦ r
The non-zero sample sequences of <N) are respectively represented by x _wr (k, r) and x _wh
We will denote it as (k, r).

In the windowing processing unit 104, as shown in FIG. 6, 179 is applied to the sample sequence x _wh (k, r) of 256 samples of one block on which the Hamming window of the above equation (10) is applied.
Two samples of 0 data are added (so-called zero padding) to make 2048 samples, and the orthogonal transformation unit 105 performs orthogonal transformation such as FFT (Fast Fourier Transform) on the time-axis data sequence of 2048 samples. Processing is performed.

In the pitch extraction unit 103, the above x _wr (k, r)
Pitch extraction is performed based on the sample sequence (1 block N samples). The pitch extraction method is known to include periodicity of time waveform, periodic frequency structure of spectrum, and autocorrelation function.
The center correlation waveform autocorrelation method is used. Regarding the center clip level in the block at this time, one clip level may be set for one block, but the peak level of the signal of each part (each sub-block) obtained by subdividing the block is detected and When there is a large difference in the peak level of each sub-block, the clip level is changed stepwise or continuously within the block. The pitch period is determined based on the peak position of the autocorrelation data of this center clip waveform. At this time, a plurality of peaks are obtained from the autocorrelation data belonging to the current frame (the autocorrelation is obtained for the data of N samples of one block), and the maximum peak among the plurality of peaks is equal to or larger than a predetermined threshold value. In the case of, the maximum peak position is set as the pitch cycle, and in other cases, the center of the pitch is a pitch range satisfying a predetermined relationship with a pitch other than the current frame, for example, the pitch before and after the frame, for example, the pitch of the previous frame. As a result, a peak in the range of ± 20% is obtained, and the pitch of the current frame is determined based on this peak position. In this pitch extraction unit 103, a relatively rough pitch search is performed by an open loop, and the extracted pitch data has a high precision (fine) pitch search unit 10.
Then, the high precision pitch search (pitch fine search) is performed by the closed loop.

High precision (fine) pitch search unit 106
Includes rough pitch data of integer (integer) values extracted by the pitch extraction unit 103 and the orthogonal transformation unit 10.
5, the data on the frequency axis subjected to FFT, for example, is supplied. In this high precision pitch search unit 106,
Centering on the above coarse pitch data value, it is ± 0.2 in increments of ±
Shake several samples at a time to drive to the optimum fine pitch data value with a decimal point (floating). As a fine search method at this time, a so-called analysis by synthesis method is used, and the pitch is selected so that the synthesized power spectrum is closest to the power spectrum of the original sound.

The fine search of the pitch will be described. First, in the MBE vocoder, the FF
Let S (j) as spectrum data on the frequency axis orthogonally transformed by T etc. be expressed as S (j) = H (j) | E (j) | 0 <j <J (11) It is assumed that the model. Where J is
Corresponding to ω _s / 4π , sampling frequency f _s = ω _s /
When 2π is, for example, 8 kHz, it corresponds to 4 kHz. In the above equation (11), spectrum data S on the frequency axis
When (j) has a waveform as shown in A of FIG. 7, H (j) shows a spectrum envelope (envelope) of the original spectrum data S (j) as shown in B of FIG. 7, and E ( j) is shown in FIG.
3 shows a spectrum of an excitation signal (excitation) which is cyclic at an equal level as shown in C of FIG. That is,
The FFT spectrum S (j) has a spectral envelope H
It is modeled as the product of (j) and the power spectrum of the excitation signal | E (j) |.

Power spectrum of the excitation signal | E (j)
Is a spectral waveform corresponding to the waveform of one band (band) for each band on the frequency axis in consideration of the periodicity (pitch structure) of the waveform on the frequency axis determined according to the pitch. It is formed by arranging it repeatedly. The waveform for one band is obtained by performing FFT by regarding the waveform obtained by adding (filling with 0) 1792 samples of 0 data to the Hamming window function of 256 samples as shown in FIG. 6 as a time axis signal. It can be formed by cutting out an impulse waveform having a certain bandwidth on the frequency axis according to the pitch.

Next, for each band divided according to the above pitch, a value (a kind of amplitude) | A that represents the above H (j) (minimizes the error for each band) | A _m
Ask for |. Here, for example, when the lower and upper points of the m-th band (band of the m-th harmonic) are a _m and b _m , respectively, the error ε _m of the m-th band is

[0039]

[Equation 1]

Can be expressed as | A _m | that minimizes this error ε _m is

[0041]

[Equation 2]

Thus, when | A _m | in this equation (13),
Minimize the error ε _m . Such an amplitude | A _m | is obtained for each band, and the obtained amplitude | A _m | is used to obtain the error ε m for each band defined by the above equation (12). Next, the sum total value Σε _m of all the bands of such error ε _m for each band is obtained. Further, such an error sum value Σε _m of all bands is obtained for some slightly different pitches, and a pitch that minimizes the error sum value Σε _m is obtained.

That is, with the rough pitch obtained by the pitch extraction unit 103 as the center, several types are prepared up and down in steps of, for example, 0.25. The error sum value Σε _m is obtained for each of these plural kinds of slightly different pitches. In this case, when the pitch is determined, the bandwidth is determined, and from the above equation (13), the power spectrum | S (j) |
(j) | can be used to find the error ε _m in the above equation (12), and the total sum Σε _m of all the bands can be obtained. This error sum value Σε _m is obtained for each pitch, and the pitch corresponding to the minimum error sum value is determined as the optimum pitch. As described above, the high-precision pitch search unit 106 obtains the optimum fine (eg, 0.25 step) pitch, and the amplitude | A _m | corresponding to this optimum pitch.
Is determined.

In the above description of the pitch fine search, in order to simplify the explanation, all bands are voiced (Vo
Assuming the case of iced), as described above, MBE
In the vocoder, unvoiced sound (Un
Since a model in which a voiced) area exists is used, it is necessary to distinguish voiced sound / unvoiced sound for each band.

The data of the optimum pitch and the amplitude | A _m | from the high precision pitch search unit 106 is sent to the voiced sound / unvoiced sound discrimination unit 107, and the voiced sound / unvoiced sound is discriminated for each band. NSR (noise to signal ratio) is used for this determination. That is, the NSR of the m-th band is

[0046]

[Equation 3]

When this NSR value is larger than a predetermined threshold value (for example, 0.3) (error is large), | A _m || E (j) | It can be judged that the approximation of | is not good (the above excitation signal | E (j) | is unsuitable as a basis), and the band is UV (Unvoice).
d, unvoiced sound). In other cases, it can be judged that the approximation has been performed to some extent, and the band is set to V
(Voiced, voiced sound).

Next, the amplitude re-evaluation unit 108 has the amplitude on the frequency axis data from the orthogonal transformation unit 105 and the amplitude | A _m evaluated as the fine pitch from the high precision pitch search unit 106.
| And each voiced sound / unvoiced sound discrimination unit 107
V / UV (voiced sound / unvoiced sound) discrimination data from The amplitude re-evaluation unit 108 re-calculates the amplitude of the band determined to be unvoiced sound (UV) by the voiced sound / unvoiced sound determination unit 107. The amplitude | A _m | _UV for this UV band is

[0049]

[Equation 4]

It is calculated by

The data from the amplitude re-evaluation unit 108 is
Data number conversion (a kind of sampling rate conversion) unit 10
Sent to 9. This data number conversion unit 109 is for making the number constant, considering that the number of divided bands on the frequency axis differs according to the pitch and the number of data (especially the number of amplitude data) differs. is there. That is, for example, when the effective band is up to 3400 Hz , the effective band is divided into 8 bands to 63 bands according to the pitch, and the amplitude | A obtained for each of these bands | A
_{The number of m m} (including UV band amplitude | A _m | _UV ) data m _MX +1 also changes from 8 to 63. Therefore, the data number conversion unit 109 converts the variable number m _MX +1 of amplitude data into a fixed number N _C (for example, 44) of data.

Here, in this embodiment, dummy data for interpolating values from the last data in the block to the first data in the block for the amplitude data of one block of the effective band on the frequency axis. Is added to expand the number of data to N _F , and then the bandwidth-limited K _OS times (for example, 8
Obtain an amplitude data of K _OS times the number by performing oversampling multiplied), the K _OS times the number ((m _MX +
1) amplitude data of × K _OS pieces) extended to linear interpolation to more NM number (e.g. 2048), the data of the N thinned out _M data the predetermined number N _C (e.g. 44) Convert to.

The data from the data number conversion unit 109 (the above-mentioned fixed number N _C of amplitude data) is sent to the vector quantization unit 110, and a predetermined number of data is collected into a vector, and vector quantization is performed. Is given. The quantized output data from the vector quantizer 110 is output to the output terminal 1
It is taken out via 11. The high-precision (fine) pitch data from the high-precision pitch search unit 106 is coded by the pitch coding unit 115, and the output terminal 11
It is taken out via 2. Further, the voiced sound / unvoiced sound (V / UV) discrimination data from the voiced sound / unvoiced sound discrimination unit 107 is taken out through the output terminal 113. The data from these output terminals 111 to 113 are transmitted as signals in a predetermined transmission format.

Note that each of these data is obtained by processing the data in the block of N samples (for example, 256 samples), but the block is a frame of L samples on the time axis. , The data to be transmitted is obtained in the frame unit. That is, the pitch data, the V / UV discrimination data, and the amplitude data are updated at the above frame cycle.

Next, a synthesizing side (decoding side) for synthesizing a voice signal on the basis of the above-mentioned respective data transmitted and obtained.
The general configuration of will be described with reference to FIG. In FIG. 8, the vector-quantized amplitude data is input to the input terminal 121, the encoded pitch data is input to the input terminal 122, and the V-value is input to the input terminal 123.
/ UV discrimination data is supplied respectively. Input terminal 12
The quantized amplitude data from 1 is the inverse vector quantization unit 12
4 and is inversely quantized, is then sent to the data number inverse transform unit 125 and is inversely transformed, and the obtained amplitude data is sent to the voiced sound synthesis unit 126 and the unvoiced sound synthesis unit 127. The encoded pitch data from the input terminal 122 is the pitch decoding unit 1
It is decoded at 28 and sent to the data number inverse conversion unit 125, the voiced sound synthesis unit 126, and the unvoiced sound synthesis unit 127. The V / UV discrimination data from the input terminal 123 is sent to the voiced sound synthesis unit 126 and the unvoiced sound synthesis unit 127.

In the voiced sound synthesis unit 126, for example, cosine (cosin)
e) A voiced sound waveform on the time axis is synthesized by wave synthesis, and in the unvoiced sound synthesis unit 127, for example, white noise is filtered by a bandpass filter to synthesize the unvoiced sound waveform on the time axis, and these voiced sound synthesized waveforms are combined. The unvoiced sound synthesized waveform is added and synthesized by the addition unit 129 and is taken out from the output terminal 130. In this case, the amplitude data, the pitch data, and the V / UV discrimination data are updated and given for each frame (L sample, for example, 160 samples) at the time of the analysis, but the continuity between the frames is improved (smoothed). Therefore, each value of the amplitude data and the pitch data is set as each data value at, for example, the center position in one frame, and each data value up to the center position of the next frame (one frame at the time of composition) is interpolated. Ask by. That is,
In one frame (for example, from the center of the above analysis frame to the center of the next analysis frame) at the time of synthesis, each data value at the tip sample point and the end (the tip of the next synthesis frame)
Each data value at the sample point is given, and each data value between these sample points is obtained by interpolation.

The synthesis processing in the voiced sound synthesis unit 126 will be described in detail below. The one combined frame (L sample, for example, 160 samples) on the time axis in the m-th band (band of the m-th harmonic) determined to be V (voiced sound)
When the voiced sound for a minute is V _m (n), V _m (n) = A _m (n) cos (θ _m (n)) 0 using the time index (sample number) n in this composite frame. ≦ n <L (16) The final voiced sound V (n) is synthesized by adding (ΣV _m (n)) the voiced sounds of all the bands which are determined to be V (voiced sound) of all the bands.

A _m (n) in the equation (16) is the amplitude of the m-th harmonic wave that is interpolated from the beginning to the end of the composite frame. The simplest way is to linearly interpolate the value of the m-th harmonic of the amplitude data updated in frame units.
That is, the m-th frame at the tip (n = 0) of the composite frame
The amplitude value of the harmonic is A _0m , the end of the composite frame (n =
L: the amplitude value of the m-th harmonic at the end of the next composite frame) is A _Lm , A _m (n) = (Ln) A _0m / L + nA _Lm / L ... (17) It suffices to calculate A _m (n).

Next, the phase θ _m (n) in the above equation (16) is given by θ _m (n) = mω _O1 n + n ² m (ω _L1 −ω ₀₁ ) / 2L + φ _{0 m} + Δω n (18) You can ask. In the equation (18), φ _0m represents the phase of the m-th harmonic (frame initial phase) at the tip (n = 0) of the composite frame, and ω ₀₁ is at the tip of the composite frame (n = 0). The fundamental angular frequency, ω _L1, represents the fundamental angular frequency at the end of the combined frame (n = L: the leading end of the next combined frame). Δω in the equation (18) is
Set a minimum Δω such that the phase φ _{Lm at} n = L is equal to θ _m (L).

On the other hand, in the embodiment of the present invention, the phase is predicted and calculated on the combining side without sending φ0m + Δωn in the equation (18) to the combining side. That is, the phase prediction unit 2
2 is obtained by adding m (ω _O1 + ω _L1 ) L / 2 to the phase (frame initial phase) ψ _0m of the m-th harmonic at time 0 (the beginning of the frame) as shown in the above equation (1). The phase ψ _{Lm at} is predicted and calculated. Further, the phase φ _{m of} each band is shown by adding ε _m to the predicted and calculated phase ψ _Lm . This ε _m indicates the prediction correction term in each band. In the present invention, Gaussian noise is used for this prediction correction term ε _m .

Here, A of FIG. 9 shows an example of a spectrum of a voice signal, and each band of band numbers (harmonics number) m is 8, 9 and 10 is UV (unvoiced sound), and other bands. Is V (voiced sound). This V
The time axis signal of the (voiced sound) band is the voiced sound synthesis unit 1 described above.
26, and the time axis signal of the UV (unvoiced sound) band is synthesized by the unvoiced sound synthesis unit 127.

The unvoiced sound synthesizing process in the unvoiced sound synthesizing section 127 will be described below. The white noise signal waveform on the time axis from the white noise generation unit 131 is windowed by a suitable window function (for example, Hamming window) with a predetermined length (for example, 256 samples), and the STFT processing unit 132 performs STFT (short circuit). By performing the (Term Fourier Transform) process, a power spectrum of white noise on the frequency axis as shown in B of FIG. 9 is obtained. The power spectrum from the STFT processing unit 132 is converted to the band amplitude processing unit 1
33, and as shown in FIG. 9C, the above-mentioned amplitude | A _m | _UV is multiplied with respect to the band (for example, m = 8, 9, 10) set as the UV (unvoiced sound), and another V (voiced sound) is generated. ) Is set to 0. This band amplitude processing unit 1
The above amplitude data, pitch data, and V / UV discrimination data are supplied to 33. The output from the band amplitude processing unit 133 is sent to the ISTFT processing unit 134, and the phase is converted into a signal on the time axis by performing inverse STFT processing using the phase of the original white noise. The output from the ISTFT processing unit 134 is sent to the overlap addition unit 135, and the overlap and addition are repeated while performing appropriate weighting (so that the original continuous noise waveform can be restored) on the time axis, and continuous. Time-domain waveforms are synthesized.
The output signal from the overlap adder 135 is sent to the adder 129.

As described above, the signals of the voiced sound portion and the unvoiced sound portion which are synthesized by the synthesis units 126 and 127 and are returned to the time axis are added by the addition unit 129 at an appropriate fixed mixing ratio, The reproduced audio signal is taken out from the output terminal 130.

Therefore, in the specific example in which the speech analysis / synthesis method according to the present invention is applied to MBE, the magnitude and variance can be controlled by making the noise used for the phase prediction Gaussian.

Regarding the configuration on the speech analysis side (encoding side) in FIG. 3 and the configuration on the speech synthesis side (decoding side) in FIG. 7, each part is described in hardware, but a so-called DSP (digital It is also possible to realize it by a software program using a signal processor or the like.

[0066]

In the voice analysis / synthesis method according to the present invention, the data on the frequency axis obtained by converting the voice signal of each block into the frequency axis is divided based on the pitch information obtained from the voice signal of each block. The analysis side obtains and transmits power information and discrimination information of voiced or unvoiced sound for each of a plurality of bands,
On the combining side, the block end phase is predicted based on the pitch information and the block initial phase for each block obtained by transmission, and the predicted end phase is a Gaussian noise having a variance according to each band. By modifying with, it is possible to control the magnitude and variance of noise, and it is expected to improve the sound quality. Further, by utilizing the signal level of the voice and its temporal change, it is possible to prevent the accumulation of errors and prevent the deterioration of the sound quality at the vowel part or at the transition point from the vowel part to the consonant part.

[Brief description of drawings]

FIG. 1 is a functional block diagram of a specific example in which a voice analysis / synthesis method according to the present invention is applied to a so-called vocoder.

FIG. 2 is a characteristic diagram for explaining Gaussian noise used in the speech analysis / synthesis method according to the present invention.

FIG. 3 is a functional block diagram showing a schematic configuration of an analysis side (encoding side) of a speech signal synthesis analysis coding apparatus as a specific example of an apparatus to which a speech analysis synthesis method according to the present invention is applied.

FIG. 4 is a diagram for explaining a windowing process.

FIG. 5 is a diagram for explaining a relationship between windowing processing and a window function.

FIG. 6 is a diagram showing time axis data as an object of orthogonal transform (FFT) processing.

FIG. 7 is a diagram showing spectrum data on a frequency axis, a spectrum envelope (envelope), and a power spectrum of an excitation signal.

FIG. 8 is a functional block diagram showing a schematic configuration on the synthesis side (decoding side) of a speech signal synthesis analysis coding apparatus as a specific example of an apparatus to which the speech analysis synthesis method according to the present invention is applied.

FIG. 9 is a diagram for explaining unvoiced sound synthesis when synthesizing voice signals.

[Explanation of Codes] 10 ... Analysis unit 20 ... Synthesis unit 21 ... Voiced sound synthesis unit 22 ... Phase prediction unit 23 ... Noise addition unit 24・ ・ ・ ・ ・ Phase correction unit 25 ・ ・ ・ Sine wave generation unit 26 ・ ・ ・ ・ ・ Amplitude amplification unit

Claims (2)

[Claims] 1. A step of dividing an input voice signal in block units to obtain pitch information within a block, and a step of converting the voice signal of each block into a frequency axis to obtain data on a frequency axis, The step of dividing the data on the frequency axis into a plurality of bands based on the pitch information, the step of obtaining the power information and the discrimination information of voiced sound or unvoiced sound for each of the divided bands, and the steps And a step of transmitting the pitch information, power information for each band, and discrimination information of voiced sound or unvoiced sound, and a block end phase based on the transmitted pitch information and block initial phase of each block. And a step of correcting the predicted block termination phase by using noise having a variance according to each band. How to do it. 2. The voice analysis / synthesis method according to claim 1, wherein the noise is Gaussian noise. [0000]