JPH06202695A - Speech signal processor - Google Patents

Abstract

PURPOSE:To improve the articulation of a synthesized speech without spoiling a feeling of naturalness by emphasizing the high-frequency formant of a frequency spectrum by directly operating parameters of the frequency range. CONSTITUTION:Quantized amplitude data from an input terminal 11 are sent to an inverse quantization 14 and quantized inversely, and the resulting data are sent to a data quantity inverse conversion part 15 and converted inversely into amplitude values by bands, which are sent to a high-frequency formant emphasis part 16. The high-frequency emphasis part 16 directly operates the amplitude values by the bands as the parameters of the frequency range to perform a high frequency formant emphasizing process and a high-frequency emphasizing process. Amplitude data obtained by the emphasizing process are sent to a voiced sound synthesis part 17 and a voiceless sound synthesis part 20. Then signals of the voiced sound part and voiceless sound part which are synthesized by the synthesis parts 17 and 20 and put back onto the time base are added by an addition part 31 at a proper fixed mixing rate and then a reproduced speech signal is outputted from an output terminal 32.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech signal processing apparatus used in a speech synthesis system, and more particularly to a post filter of a speech synthesis system of a speech decoding apparatus of multi-band excitation coding (MBE). The present invention relates to a preferred audio signal processing device.

[0002]

2. Description of the Related Art Various coding methods are known in which signal compression is performed by utilizing the statistical characteristics of a voice signal in the time domain and frequency domain and the characteristics of human hearing. This speech coding method is roughly classified into coding in the time domain,
Examples include coding in the frequency domain and synthesis analysis coding.

Specific examples of the coding of the voice signal include MBE (Multiband Excitation) coding, SBE (Singleband Excitation) coding, Harmonic coding, and SBC (Multiband Excitation). Sub-band Coding), LP
There are C (Linear Predictive Coding), DCT (Discrete Cosine Transform), MDCT (Modified DCT), FFT (Fast Fourier Transform), and the like.

[0004]

By the way, the above MBE
In a speech analysis / synthesis system such as coding, which mainly involves processing in the frequency domain, quantization distortion causes spectral distortion, and deterioration in the high frequency range, where bit allocation is usually small, becomes significant. Many. As a result, the speech synthesized from this spectrum loses clarity due to the loss of high-range formants or lack of power, and lack of power in the entire high range, so that a so-called stuffy nose can be heard. Is becoming.

To correct this, a formant emphasizing filter using, for example, an IIR (infinite impulse response) filter for performing the compensation processing in the time domain was used, but in this case, a speech processing frame is used. The filter coefficient for emphasizing the formant must be calculated for each time, and real-time processing is difficult. Also, it is necessary to pay attention to the stability of the filter, and there is a drawback that the effect is not great for the amount of calculation processing.

The present invention has been made in view of the above circumstances, and provides a speech signal processing apparatus in which processing such as formant enhancement in a speech synthesis system is simplified and real-time processing can be easily performed. To aim.

[0007]

SUMMARY OF THE INVENTION In order to solve the above-mentioned problems, a voice signal processing device according to the present invention is a voice signal processing device used in a voice synthesizing system centering on processing in a frequency domain. The feature is that the high-band formant of the spectrum is emphasized by directly operating the parameters in the frequency domain.

Here, it is conceivable to emphasize not only the high-frequency formant of the frequency spectrum but also the high-frequency band by directly operating the parameters in the frequency domain. The speech signal processing device having such characteristics is preferably applied to a post-filter of a speech synthesis system of a multiband excitation coding (MBE) type speech decoding device.

[0009]

Since the parameter is directly operated in the frequency domain, a high-frequency emphasis filter (eg IIR) in the time axis direction is used.
Without using a filter or the like, only a portion to be emphasized can be accurately emphasized with a simple configuration and a simple operation, and real-time processing can be easily performed.

[0010]

Embodiments of the audio signal processing apparatus according to the present invention will be described below with reference to the drawings. FIG. 1 shows a schematic configuration of a multi-band excitation coding (MBE) type speech decoding apparatus as a speech synthesis system to which a speech signal processing apparatus according to an embodiment of the present invention is applied.

The input terminal 11 shown in FIG.
Quantized amplitude data transmitted from a BE type speech encoding device, a so-called MBE vocoder, is supplied.
In the MBE vocoder, the quantized amplitude data is obtained by dividing the spectrum of each processing frame of the input voice signal by the pitch of the voice signal as a unit, and determining the amplitude value for each band by a fixed number of data independent of the pitch value. Converted to
This is the data obtained by vector quantization. Input terminal 1
Pitch data encoded by the MBE vocoder and V / UV discrimination data indicating voiced sound or unvoiced sound for each band are supplied to 2 and 13, respectively.

The quantized amplitude data from the input terminal 11 is sent to the inverse vector quantizer 14 and inversely quantized,
After being sent to the data number inverse conversion unit 15 and inversely converted into the amplitude value for each band, it is sent to the high-frequency formant emphasizing unit 16 which is a main part of the embodiment of the present invention. The high-frequency formant emphasizing unit 16 performs high-frequency formant emphasizing processing and high-frequency emphasizing processing by directly operating the amplitude value for each band, which is a parameter in the frequency domain. The amplitude data obtained by this emphasis processing is sent to the voiced sound synthesis unit 17 and the unvoiced sound synthesis unit 20.

The encoded pitch data from the input terminal 12 is decoded by the pitch decoding unit 18 and sent to the data number inverse conversion unit 15, voiced sound synthesis unit 17 and unvoiced sound synthesis unit 20. The V / UV discrimination data from the input terminal 13 is sent to the voiced sound synthesis unit 17 and the unvoiced sound synthesis unit 20. The voiced sound synthesis unit 17 synthesizes a voiced sound waveform on the time axis by, for example, cosine wave synthesis and sends it to the addition unit 31.

In the unvoiced sound synthesizing unit 20, first, the white noise signal waveform on the time axis from the white noise generating unit 21 is windowed with an appropriate window function (eg, Hamming window) with a predetermined length (eg, 256 samples). Call S
By performing STFT (Short Term Fourier Transform) processing by the TFT processing unit 22, a power spectrum on the frequency axis of the white noise signal is obtained. This STFT
The power spectrum from the processing unit 22 is sent to the band amplitude processing unit 23, and the band of UV (unvoiced sound) is multiplied by the amplitude | A _m | _UV to determine the amplitude of another V (voiced sound) band. Set to 0. The band amplitude processing unit 23 is supplied with the amplitude data, the pitch data, and the V / UV discrimination data. The output from the band amplitude processing unit 23 is sent to the ISTFT processing unit 24, 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 24 is sent to the overlap adding unit 25, which repeats overlap and addition while appropriately weighting (so that the original continuous noise waveform can be restored) on the time axis, and continuously. Time-domain waveforms are synthesized. The output signal from the overlap adder 25 is sent to the adder 31.

As described above, the signals of the voiced sound portion and the unvoiced sound portion which are synthesized in the respective synthesis units 17 and 20 and returned on the time axis are added by the addition unit 31 at an appropriate fixed mixing ratio. , The reproduced audio signal is taken out from the output terminal 32.

Next, the high-frequency formant emphasis processing and high-frequency emphasis processing in the high-frequency formant emphasis unit 16 will be described with reference to the flowchart of FIG. The amplitude value for each band, which is the spectrum information from the data number inverse conversion unit in FIG. 1, that is, the amplitude value for each band obtained by frequency-dividing the spectrum of the input voice signal on the MBE vocoder side in accordance with the pitch, Let A (n). Here, n is a so-called harmonic number or band index number, which is an integer value incremented by the pitch period on the frequency axis.

First, in step S1 of FIG. 2, the maximum value of the amplitude values A (n) for each band is detected. In the next step S2, it is determined whether or not this maximum value is larger than a predetermined threshold value J ₁ , and when the result is NO, the high-frequency formant emphasis process is not performed and the process ends. If YES, the process proceeds to step S3 to start the high frequency formant emphasis process as described below. This is because when the spectrum value (amplitude value for each band) is originally small, it is rather unnatural when emphasis is performed, and only when the maximum value is larger than the threshold value J _1. The following emphasis processing is performed.

Here, as a specific example for explaining the processing for emphasizing the voice formant, when the amplitude value A (n) for each band is as shown in FIG. 3, for example, F in the figure is used. ₁ and F ₂ clearly show the shape of the peak of the mountain, and it can be seen that these are the first and second formants of the voice, but the portion shown in F ₃ has the shape It is unclear whether it is a formant or not. Therefore, these A (n) envelopes are representatively or estimated by the moving average of A (n), and A ((n) is represented with respect to the value A '(n) of each envelope representative or estimated by this moving average. n) is larger or smaller,
Formant enhancement is performed by making it larger when it is large and smaller when it is small.

That is, in step S3 of FIG. 2, the moving average A '(n) of the amplitude value A (n) for each band is

[0020]

[Equation 1]

It is calculated by

B in the equation (1) represents the total bandwidth for which the moving average is to be taken, n _max represents the maximum value among the determined frequencies, and changes with the angular frequency ω _{1 of the} pitch. Is.

FIG. 4 shows an example in which the amplitude value A (n) and the moving average A '(n) for each band are plotted. This Figure 4
In, the buried formant F ₃ on the high frequency side is emphasized by the following expression (2). Ae (n) = A (n) + f (A (n) -A '(n)) ・ a (n) (2)

In the equation (2), Ae (n) is an emphasis of the formant of the original spectrum (amplitude value for each band) A (n). That is, when the original spectrum is larger than the envelope obtained by the moving average, the original spectrum is further increased, and in the opposite case, the operation is further reduced. Further, as the function f in the above formula (2), specifically, for example, a function like the following formula (3) may be used.

[0025]

[Equation 2]

In the equation (3), sgn (p) is a function indicating the sign (positive or negative polarity) of p, and K is a constant for saturation when p exceeds a certain value K. is there.
The function of equation (3) is shown in FIG.

Further, a (n) in the above equation (2) is data for changing the degree of emphasis depending on the band, and is set to 0 or an extremely small value on the low frequency side and non-zero on the high frequency side. The value of should be emphasized.

However, when the amplitude value A (n) itself, which is assumed to increase Ae (n) after the emphasis, becomes equal to or larger than a predetermined threshold value J ₂ , the above emphasis is not performed.
Therefore, in step S4 of FIG. 2, it is determined whether or not the amplitude value A (n) is smaller than the threshold value J ₂ , and only when YES, the process proceeds to step S5 to calculate the equation (2) and emphasize it. After obtaining the calculated amplitude value Ae (n), the process proceeds to step S7,
If NO, the process proceeds to step S6 and the original amplitude value A (n)
As Ae (n) (without emphasizing the above) step S7
Proceed to.

In addition to the high range formant enhancement as described above, in order to correct the overall power shortage in the high range, the following formula (4) is used based on preset high range enhancement data. The high frequency component is increased to enhance the high frequency range. Aee (n) = Ae (n) + em (n) (4) em (n) in the equation (4) is a data string that has information about how much emphasis is to be applied to band n. is there.

With regard to this high-frequency emphasis, when the voice pitch to be analyzed is high (female voice, etc.), the number of bands in the frequency band to be analyzed (for example, 200 to 3400 Hz) is small,
Considering that the error of the whole spectrum is also small, the above e
Set m (n) to a smaller value, for example half the value em (n) / 2. That is, it is determined in step S7 whether or not the value Np representing the voice pitch of the analysis target in the number of samples is smaller than a predetermined threshold value P. If YES, the process proceeds to step S8, where em (n) = em (n ) / 2 and proceed to step S9,
If NO, the process directly proceeds to step S9. The pitch threshold value P in this case may be, for example, P = 40 (corresponding to a pitch of 200 Hz at 8 kHz sampling). When Np <40, the voice pitch is higher than 200 Hz. In step S9, the high range emphasis of the above formula (4) is performed.

When performing high-frequency formant emphasis or high-frequency emphasis as described above, unnatural sound may occur depending on the level after emphasis, and in order to avoid this, the above-mentioned steps S2 and S4 and steps S7 and S8
The following measures are taken. Further, when the high-frequency formant is emphasized, the band to be emphasized may be limited, and for example, it may be set in advance so that the formant emphasis is not carried out on the band having the harmonics number n of 12 or less. This band limiting process is performed by using a (n) in the above equation (2).
It may be performed by the function of.

Next, a specific example of a kind of MBE (Multiband Excitation) vocoder of a voice signal synthesis analysis coding apparatus (so-called vocoder) to which the voice signal processing apparatus according to the present invention can be applied will be described with reference to the drawings. It will be explained with reference to FIG. This MBE vocoder is based on DW Griffi
n and JS Lim, Multiband Excitation Vocoder, "IEE
E Trans. Acoustics, Speech, and Signal Processing, vo
L.36, No.8, pp. 1223-1235, Aug.1988, the conventional PARCOR (PARtialauto-CORr
elation: Partial autocorrelation) In a vocoder or the like, a voiced sound section and an unvoiced sound section were switched for each block or frame when modeling speech, whereas in the MBE vocoder, the same time (in the same block or frame) Voiced section and unvoiced sound (Unvoice
d) It is modeled on the assumption that an interval and exists.

FIG. 6 is a block diagram showing an overall schematic configuration of the embodiment of the MBE vocoder. In FIG. 6, an audio signal is supplied to an input terminal 101, and this input audio signal is sent to a filter 102 such as an HPF (high-pass filter) to be a so-called DC signal.
Removal of (DC) offset and band limitation (for example, 200
At least low frequency component (2)
(Less than 00 Hz) is removed. The signal obtained through the filter 102 is sent to the pitch extraction unit 103 and the windowing processing unit 104, respectively. Pitch extraction unit 103
In, the input voice signal data is divided into blocks in units of a predetermined number N (for example, N = 256) (or cut out by a rectangular window), and pitch extraction is performed on the voice signals in this block. Such a cut block (256 samples) is moved in the time axis direction at a frame interval of, for example, L samples (for example, L = 160), and the overlap between blocks is NL samples (for example, 96 samples). Has become. In addition, the windowing processing unit 104 applies a predetermined window function, for example, a Hamming window, to one block of N samples, and sequentially moves the windowed block in the time axis direction at intervals of one frame of L samples. The orthogonal transform unit 105 performs an orthogonal transform process such as FFT (Fast Fourier Transform) on the data string of the output signal subjected to the windowing process.

The pitch extraction unit 103 determines the peak period by using, for example, the auto-correlation method 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 of 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 pitch is within the 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.
6, the pitch search (pitch fine search) with high accuracy 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, for example, FFT-processed data on the frequency axis is supplied. In this high precision pitch search unit 106,
Centering on the above coarse pitch data value, ± in increments of 0.2 to 0.5
Shake several samples at a time to reach 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 optimum pitch and amplitude | A _m | data 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, this NSR
When the value is larger than a predetermined threshold value (for example, 0.3) (large error), the band is determined to be UV (Unvoiced, unvoiced sound). In other cases, it can be determined that the approximation has been performed to some extent, and that band is set to V (Voice
d, voiced sound).

Next, in the amplitude re-evaluation unit 108, 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 is supplied. The amplitude re-evaluation unit 108 again obtains the amplitude | A _m | _UV for the band determined to be unvoiced sound (UV) by the voiced sound / unvoiced sound determination unit 107.

The data from the amplitude re-evaluation unit 108 is
Data number conversion (a kind of sampling rate conversion) unit 10
Sent to 9. The data number conversion unit 109 is for making the number constant in consideration of the fact 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} | (including UV band amplitude | A _m | _UV ) data also varies from 8 to 63. Therefore, the data number conversion unit 109 converts the variable number 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 of one block of the effective band on the frequency axis. Is added to increase the number of data to N _F , and then the bandwidth-limited K _OS times (for example, 8
By multiplying the number of K _OS times the amplitude data, and multiplying the number of K _OS times ((m _MX +
1) × K _OS pieces of amplitude data are linearly interpolated to be expanded to a larger number of N _M pieces (for example, 2048 pieces), and the N _M pieces of data are thinned out to obtain the fixed number N _C (for example, 44 pieces). Convert to data.

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 is grouped into a vector for each predetermined number of data, 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. Further, 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), and the block is a frame of the 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 period.

Regarding the configuration of the voice analysis side (encoding side) of FIG. 6 and the configuration of the voice synthesis side (decoding side) of FIG. 1, although each unit is described in hardware, a so-called DSP (digital) is used. It is also possible to realize it by a software program using a signal processor or the like.

[0043]

According to the voice signal processing device of the present invention, in the voice signal processing device used in the voice synthesizing system centering on the processing in the frequency domain, the high-frequency formant of the frequency spectrum is converted into the parameter in the frequency domain. Since it is directly operated to emphasize, only a portion to be emphasized can be accurately emphasized with a simple configuration and a simple operation, and the clarity of the synthesized sound can be improved without impairing the natural feeling. Furthermore, a high-frequency emphasis filter (for example, I
Since it is not necessary to calculate the position of the pole of the filter, which is indispensable when processing in the time domain using an (IR filter), etc., real-time processing can be performed easily, and the filter is unstable. There is an advantage that adverse effects due to the size can be completely avoided.

[Brief description of drawings]

FIG. 1 is a functional block diagram showing a schematic configuration of a speech decoding apparatus on a synthesis side (decoding side) of a speech synthesis analysis coding apparatus as a specific example of an apparatus to which an embodiment of a speech signal processing apparatus according to the present invention can be applied. It is a figure.

FIG. 2 is a flow chart for explaining the operation of the above embodiment.

FIG. 3 is a diagram showing an amplitude value for each band which is spectrum data on a frequency axis.

FIG. 4 is a spectrum envelope (envelope) obtained by taking spectrum data on a frequency axis and a moving average thereof.
FIG.

FIG. 5 is a diagram showing a function of an emphasis method at the time of formant emphasis.

FIG. 6 is a functional block diagram showing a schematic configuration of an analysis side (encoding side) of a speech synthesis analysis coding apparatus which sends a signal to a speech decoding apparatus to which the embodiment of the speech signal processing apparatus according to the present invention is applied. .

[Explanation of symbols]

 11: Quantized amplitude data input terminal 12: Encoded pitch data input terminal 13: V / UV discrimination data input terminal 16: High frequency formant emphasis section

Claims (3)

[Claims] 1. A speech signal processing device used in a speech synthesis system, which is mainly processing in the frequency domain, wherein a high-frequency formant of a frequency spectrum is emphasized by directly operating a parameter in the frequency domain. Signal processing device. 2. The audio signal processing apparatus according to claim 1, wherein the entire high frequency band of the frequency spectrum is emphasized by directly operating parameters in the frequency domain. 3. The speech signal processing device according to claim 1, wherein the speech synthesis system is a speech synthesis system of a speech decoding device for multi-band excitation coding.