JP2001142499A - Speech encoding device and speech decoding device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a speech encoding device and a speech decoding device which can obtain a satisfactory tone quality of encoding regardless of a low bit rate with respect to speech on which a background noise is superposed. SOLUTION: A discrimination part 800 extracts a feature quantity from a sound signal to discriminate a mode, and a smoothing circuit 450 smoothes at least one of the gain, of a sound source signal, the gain of an adaptive code book, a spectrum parameter, and the level of the sound source signal in the time direction if the discrimination output is a preliminarily determined mode. A multiplexer part 400 uses the smoothed signal to locally reproduce a synthesized signal and combines and outputs the output of a spectrum parameter calculation part, that of the discrimination part, that of an adaptive code book part, that of a sound source quantization part, and that of a gain quantization part.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech coding apparatus and a speech decoding apparatus for satisfactorily coding a background noise signal superimposed on a speech signal even at a low bit rate.

[0002]

2. Description of the Related Art As a method for encoding a speech signal with high efficiency, for example, "Code by M. Schroeder and B. Atal"
-excited linear prediction: High quality speech at
verylow bitrates "(Proc. ICASSP, pp.937-940, 1985
), And "Imp" by Kleijn et al.
roved speech quality and efficient vector quantiza
tion in SELP "(Proc. ICASSP, pp.155-158, 1988)
CELP (Co
de Excited Linear Predictive Coding) is known.

In this conventional example, on the transmitting side, every frame
At time (for example, 20 ms), a linear parameter (LPC) analysis is used to extract a spectral parameter representing a spectral characteristic of the audio signal from the audio signal. The frame is further divided into subframes (e.g., 5 ms), and the parameters in the adaptive codebook (delay parameters and gain parameters corresponding to the pitch period) are extracted based on the past sound source signals for each subframe, and the adaptive codebook Pitch prediction of the audio signal of the subframe. For the sound source signal obtained by pitch prediction, by selecting an optimum sound source code vector from a sound source cobook (vector quantization codebook) composed of a predetermined type of noise signal and calculating an optimum gain, Quantize the sound source signal. The excitation code vector is selected so as to minimize the error power between the signal synthesized from the selected noise signal and the residual signal. Then, the index and gain indicating the type of the selected code vector, the spectrum parameter and the parameter of the adaptive codebook are combined and transmitted by the multiplexer unit. Description on the receiving side is omitted.

[0004]

In the conventional method, when the encoding bit rate is reduced to, for example, 8 kb / s or less, the background noise signal deteriorates, particularly when the background noise signal is superimposed on the audio signal. There was a problem that the overall sound quality deteriorated. This was particularly noticeable when using voice coding on a mobile phone or the like. In the methods of the above-mentioned Documents 1 and 2, when the bit rate is reduced, the number of bits of the sound source codebook is reduced, and the waveform reproduction accuracy is reduced. When the correlation between the waveforms is high as in an audio signal, the deterioration is not so remarkable, but when the signal has a low correlation such as background noise, the deterioration becomes remarkable.

[0005] Also, "16 kbps wideban" by C. Laflamme
d speech coding technique basedon algebraic CELP "
In the conventional method of the paper entitled "Proc. ICASSP, pp. 13-16, 1991" (Reference 3), since the sound source signal is represented by a combination of pulses, the consistency of the model is high and good for speech. However, when the bit rate is low, the number of pulses is insufficient, so that the sound quality of the background noise portion of the coded speech is extremely deteriorated.

The reason for this is that in a vowel section of a voice,
Since the pulses are concentrated near the pitch pulse, which is the starting point of the pitch, the number of pulses can be efficiently represented by a small number of pulses. However, for random signals such as background noise, the pulses need to be randomly generated. Therefore, it is difficult to satisfactorily represent the background noise with a small number of pulses, and when the bit rate is reduced and the number of pulses is reduced, the sound quality with respect to the background noise is rapidly deteriorated.

SUMMARY OF THE INVENTION It is therefore an object of the present invention to solve the above-mentioned problems, and to provide a speech coding apparatus and a speech decoding apparatus which have a relatively small amount of operation even when the bit rate is low, and in which the sound quality is not particularly deteriorated due to background noise. It is to provide a device.

[0008]

In order to solve the above-mentioned problems, a speech encoding apparatus and a speech decoding apparatus according to the present invention employ the following characteristic configurations.

(1) A spectrum parameter calculation unit for inputting a speech signal and obtaining and quantizing a spectrum parameter for each predetermined frame, dividing the frame into a plurality of subframes, and An adaptive codebook section for obtaining a delay and a gain by an adaptive codebook from a quantized sound source signal, predicting an audio signal to obtain a residual, and quantizing and outputting the sound source signal of the audio signal using the spectrum parameter. In a speech coding apparatus having a sound source quantization unit that performs quantization and a gain quantization unit that quantizes the gain of the adaptive codebook and the gain of the sound source signal, a predetermined feature amount is extracted and extracted from the sound signal. A mode discrimination unit for discriminating which mode among a plurality of predetermined modes based on the characteristic amount; A smoothing circuit that smoothes at least one of a gain of the excitation signal, a gain of the adaptive codebook, the spectrum parameter, and a level of the excitation signal in a time direction when an output of the unit is in a predetermined mode. Local reproduction of the synthesized signal using the smoothed signal, the output of the spectrum parameter calculation unit, the output of the determination unit, the output of the adaptive codebook unit, and the output of the sound source quantization unit. And a multiplexer unit that combines and outputs the output of the gain quantization unit.

(2) The speech coding apparatus according to (1), wherein the mode discriminating circuit discriminates a mode for each frame.

(3) The speech encoding apparatus according to (1), wherein the feature quantity is a pitch prediction gain.

(4) The mode discriminating circuit averages the pitch prediction gain obtained for each sub-frame for the entire frame, compares this value with a plurality of predetermined thresholds, and determines a plurality of predetermined thresholds. The speech encoding device according to the above (1), which is classified into modes.

(5) The speech coding apparatus according to (1), wherein the plurality of predetermined modes substantially correspond to unvoiced sections, transient sections, weak voiced sections, and strongly voiced sections.

(6) A demultiplexer circuit for inputting and separating a spectrum parameter, a pitch, a gain and a sound source signal as voice information, and a sound source signal restoring for restoring a sound source signal from the separated pitch, sound source signal and gain. A circuit, a synthesis filter circuit that synthesizes an audio signal using the restored sound source signal and the spectrum parameter,
And a post-filter circuit configured to input the synthesized speech and perform post-filtering using the spectrum parameter.In the speech decoding apparatus, inverse post-filtering and inverse synthesis filtering are performed based on an output signal of the post-filter circuit and the spectrum parameter. An inverse filter circuit for estimating a source signal to be performed, a smoothing circuit for performing time-direction smoothing on at least one of the level of the estimated source signal, the gain, and the spectral parameter, and the smoothed signal. Is input to the synthesis filter circuit, and the obtained synthesized signal is input to the post-filter circuit to synthesize an audio signal.

(7) A demultiplexer circuit for inputting and separating discrimination information indicating a mode determined based on a feature amount of a speech signal to be decoded, a spectrum parameter, a pitch, a gain, and an excitation signal, and A sound source signal restoring circuit for restoring a sound source signal from a pitch, a sound source signal, and a gain, a synthesis filter circuit for synthesizing an audio signal using the restored sound source signal and the spectrum parameter, and And a post-filter circuit that performs post-filtering using a parameter. In the case where the discrimination information is in a predetermined mode, inverse post-filtering is performed based on an output signal of the post-filter circuit and the spectrum parameter. And perform inverse synthesis filtering to estimate the sound source signal. An inverse filter circuit, a smoothing circuit for performing time-direction smoothing on at least one of the level of the estimated sound source signal, the gain, and the spectral parameter, and the synthesis filter circuit for filtering the smoothed signal. And a speech decoding device for inputting the obtained synthesized signal to the post-filter circuit to synthesize a speech signal.

(8) The speech encoding apparatus according to (7), wherein the mode determination is performed for each frame.

(9) The speech encoding apparatus according to the above (7), wherein the feature amount is a pitch prediction gain.

(10) In the mode discrimination, a pitch prediction gain obtained for each sub-frame is averaged over the entire frame,
The speech encoding device according to the above (7), wherein the value is compared with a plurality of predetermined threshold values.

(11) The speech encoding apparatus according to (7), wherein the mode substantially corresponds to an unvoiced section, a transient section, a weak voiced section, and a strong voiced section.

(12) Locally reproducing a synthesized speech signal based on a signal obtained by smoothing at least one of a spectrum parameter, a gain of an adaptive codebook, a gain of a sound source codebook, and an RMS of a sound source signal in a time direction. Speech encoding device.

(13) On the decoding side of the audio signal, a residual signal is obtained from the post-filtered signal by inverse post-synthesis filter processing, and the RMS of the residual signal, the received spectrum parameters, the gain of the adaptive codebook,
An audio decoding device that re-synthesizes an audio signal based on a signal obtained by smoothing at least one of the gains of a sound source codebook in the time direction, performs post-filtering processing again, and outputs a final synthesized signal.

(14) On the decoding side of the audio signal, a residual signal is obtained from the post-filtered signal by inverse post-synthesis filter processing, and a mode determined based on the characteristic amount of the audio signal to be decoded, or In the case where the feature amount is in a predetermined region, a signal obtained by smoothing at least one of the RMS of the residual signal, the reception spectrum parameter, the gain of the adaptive codebook, and the gain of the excitation codebook in the time direction. An audio decoding device that re-synthesizes an audio signal based on the signal, and further performs post-filtering processing to output a final synthesized signal.

[0023]

FIG. 1 is a block diagram showing a first embodiment of a speech coding apparatus according to the present invention. In FIG. 1, an audio signal is input from an input terminal 100, a frame division circuit 110 divides the audio signal into frames (for example, 20 ms), and a subframe division circuit 120 divides the audio signal of the frame into subframes shorter than the frame. (For example, 5m
s).

In the spectrum parameter calculation circuit 200,
At least one subframe audio signal is cut out over a window (for example, 24 ms) longer than the subframe length, and spectrum parameters are calculated for a predetermined order (for example, P = 10th order). Here, the well-known LPC analysis, Burg analysis, or the like can be used for calculating the spectrum parameters. Here, Burg analysis is used. Details of the Burg analysis are described in a book entitled "Signal Analysis and System Identification" by Nakamizo (Corona Publishing Co., 1988), pp. 82-87 (Reference 4), and will not be described here. Further, the spectrum parameter calculation unit 200 converts the linear prediction coefficients αi (i = 1,..., 10) calculated by the Burg method into LSP parameters suitable for quantization and interpolation. Here, the conversion from linear prediction coefficients to LSP is described in a paper entitled “Speech Information Compression by Line Spectrum Pair (LSP) Speech Analysis / Synthesis Method” by Sugamura et al. (Transactions of IEICE, J64-A, pp.599 -606, 1981) (Reference 5)
Can be referred to. For example, the linear prediction coefficients obtained by the Burg method in the second and fourth sub-frames are converted into LSP parameters, the LSPs of the first and third sub-frames are obtained by linear interpolation, and the LSPs of the first and third sub-frames are inverted. The conversion is returned to the linear prediction coefficient, and the linear prediction coefficient αil (i = 1,..., 10, l = 1,..., 5) of the first to fourth subframes is output to the audibility weighting circuit 230. Further, it outputs the LSP of the fourth subframe to spectrum parameter quantization circuit 210.

The spectrum parameter quantization circuit 210 efficiently quantizes the LSP parameter of a predetermined subframe and outputs a quantization value for minimizing the distortion of the following equation.

(Equation 1)

Here, LSP (i), QLSP (i) j, and W (i) are an i-th LSP before quantization, a j-th result after quantization, and a weight coefficient, respectively.

In the following, it is assumed that vector quantization is used as a quantization method, and that the LSP parameter of the fourth subframe is quantized. A well-known method can be used for the vector quantization of the LSP parameter. A specific method is described in, for example, JP-A-4-171500 (Japanese Patent Application No.
97600) (Reference 6) and JP-A-4-363000 (Japanese Patent Application No. Hei.
3-261925) (Reference 7), JP-A-5-6199 (Japanese Patent Application No. 3-155049) (Reference 8), and "LSP Coding Using VQ-SVQ With Interpolation" by T. Nomura et al. i
n 4.075 kbps M-LCELP Speech Coder "(Pro
c. Mobile Multimedia Communications, pp.B.2.5, 199
3) Since (Reference 9) can be referred to, the description is omitted here.

The spectrum parameter quantization circuit 21
In 0, the LSP parameters of the first to fourth subframes are restored based on the LSP parameters quantized in the fourth subframe. Here, the LSPs of the first to third subframes are restored by linearly interpolating the quantization LSP parameter of the fourth subframe of the current frame and the quantization LSP of the fourth subframe of the previous frame. Here, after selecting one type of code vector that minimizes the error power between the LSP before quantization and the LSP after quantization, the LSPs of the first to fourth subframes can be restored by linear interpolation. In order to further improve the performance, after selecting a plurality of code vectors for minimizing the error power, the cumulative distortion is evaluated for each candidate, and a combination of the candidate for minimizing the cumulative distortion and the interpolation LSP is determined. Can be selected. For details, see, for example, JP-A-6-222797 (Japanese Patent Application No. 5-8737) (Reference 10).
Can be referred to.

The LSPs of the first to third sub-frames and the quantized LSP of the fourth sub-frame, which have been reconstructed as described above, are assigned to the linear prediction coefficients α _il (i = 1,..., 10, l = 1,. ) And outputs the result to the impulse response calculation circuit 310. Further, an index representing the code vector of the quantized LSP of the fourth subframe is output to the multiplexer 400.

The perceptual weighting circuit 230 inputs the linear prediction coefficient α _il (i = 1,..., 10, l = 1,..., 5) from the spectrum parameter calculation circuit 200 for each subframe before quantization. Based on Document 1, perceptual weighting is performed on the audio signal of the subframe, and a perceptual weighting signal is output.

The response signal calculation circuit 240 receives the linear prediction coefficient α _il for each subframe from the spectrum parameter calculation circuit 200 and
From 0, a linear prediction coefficient α _il restored by quantization and interpolation is input for each subframe, and the response signal is set to 0 d (n) = 0 using the saved filter memory value. Is calculated for one subframe and output to the subtractor 235. Here, the response signal xz (n) is represented by the following equation.

(Equation 2)

Here, N indicates a subframe length. γ
Is a weighting coefficient for controlling the perceptual weighting amount, and has the same value as the following equation (7). sw (n) and p (n) are
The output signal of the weighting signal calculation circuit and the output signal of the denominator term of the filter of the first term on the right side in equation (7) described below are shown.

The subtractor 235 subtracts the response signal for one subframe from the auditory sensation weighted signal by the following equation, and obtains x ′ w (n)
Is output to the adaptive codebook circuit 300.

(Equation 3)

The impulse response calculation circuit 310 calculates the impulse response of the auditory weighting filter whose z-transform is expressed by the following equation.
hw (n) is calculated by a predetermined number L and output to the adaptive codebook circuit 500 and the sound source quantization circuit 350.

(Equation 4)

The mode discriminating circuit 800 uses the output signal of the frame dividing circuit to extract a characteristic amount and discriminate the mode for each frame. Here, as a feature, a pitch prediction gain can be used. The pitch prediction gain obtained for each subframe is averaged for the entire frame, this value is compared with a plurality of predetermined thresholds, and the mode is classified into a plurality of predetermined modes. Here, as an example, it is assumed that the number of modes is four. In this case, mode 0, 1, 2, 3
Respectively correspond approximately to an unvoiced section, a transient section, a weak voiced section, and a strong voiced section. The mode determination information is output to the sound source quantization circuit 350, the gain quantization circuit 365, and the multiplexer 400.

In the adaptive code book circuit 500, the past sound source signal v (n) from the gain quantization circuit 370 is subtracted by the subtractor 2
An output signal x ′ w (n) is input from the block 35 and an auditory weighting impulse response hw (n) is input from the impulse response calculation circuit 310. The delay T corresponding to the pitch is determined so as to minimize the distortion of the following equation, and the index representing the delay is calculated by the multiplexer 4.
Output to 00.

(Equation 5) In equation (8), the symbol * represents a convolution operation.

Next, the gain β is obtained according to the following equation.

(Equation 6)

Here, in order to improve the accuracy of delay extraction for female sounds and children's voices, the delay may be determined not by integer samples but by decimal sample values. A specific method is described in, for example, "Pitch pre-dictors" by P. Kroon et al.
with high temporal resolution "(Proc.
ICASSP, pp.661-664, 1990) (Reference 11).

Further, in the adaptive codebook circuit 500, pitch prediction is performed according to the equation (10), and the prediction residual signal ew (n)
Is output to the sound source quantization circuit 355.

(Equation 7)

The sound source quantization circuit 355 inputs mode discrimination information, and switches the quantization method of the sound source signal depending on the mode.

In modes 1, 2, and 3, M pulses are applied. In modes 1, 2, and 3, it is assumed that a B-bit amplitude codebook or a polarity codebook for collectively quantizing the pulse amplitude for M pulses is provided. Hereinafter, a description will be given of a case where the polarity codebook is used. This polarity codebook is stored in the sound source codebook 351.

In voiced, the sound source quantization circuit 355 reads out each polarity code vector stored in the code book 351, assigns a position to each code vector, and determines a code vector and a position for minimizing the equation (11). Select multiple combinations.

(Equation 8)

Here, hw (n) is an auditory weighting impulse response. In order to minimize the expression (11), a combination of the polarity code vector g _ik and the position mi that maximizes the expression (12) may be obtained.

(Equation 9)

Alternatively, selection may be made so as to maximize equation (13). This reduces the amount of calculation required for calculating the numerator.

(Equation 10)

Here, the possible positions of each pulse in modes 1 to 3 can be constrained as shown in Reference 3 to reduce the amount of calculation. As an example, if N = 40 and M = 5, the possible positions of each pulse are as shown in Table 1.

[Table 1]

After the search for the polarity code vector is completed, the combination of the selected plurality of sets of the polarity code vector and the position is output to the gain quantization circuit 370. In a predetermined mode (mode 0 in this example), as shown in Table 2,
The positions of the pulses are determined at regular intervals, and a plurality of shift amounts for shifting the position of the entire pulse are determined. In the following example, assuming that the position is shifted one sample at a time, four types of shift amounts (shift 0, shift 1, shift 2,
Use shift 3). In this case, the shift amount is quantized by 2 bits and transmitted.

[Table 2] The polarity for each shift amount and each pulse position in Table 2 is expressed by an equation
Determine in advance from (14).

The positions shown in Table 2 and the corresponding polarities are output to the gain quantization circuit 365 for each shift amount.

The gain quantization circuit 370 receives the mode discrimination information from the mode discrimination circuit 800. In modes 1 to 3, combinations of a plurality of sets of polarity code vectors and pulse positions are input from the sound source quantization circuit 355, and in mode 0, combinations of pulse positions and corresponding polarities are input for each shift amount.

The gain quantization circuit 370 reads the gain code vector from the gain code book 380, and in the modes 1 to 3, minimizes the expression (15) with respect to the selected combination of a plurality of sets of polarity code vectors and positions. A gain code vector is searched so as to optimize the gain code vector, and one kind of combination of the gain code vector, the polarity code vector, and the position for minimizing distortion is selected.

[Equation 11]

Here, an example has been described in which both the gain of the adaptive codebook and the gain of the sound source expressed in pulses are simultaneously vector-quantized. The index representing the selected polarity code vector, the code representing the position, and the index representing the gain code vector are output to the multiplexer 400.

When the discrimination information is mode 0, a plurality of shift amounts and polarities corresponding to respective positions in the case of each shift amount are input, a gain code vector is searched, and the equation (16) is minimized. One type of gain code vector and shift amount is selected.

(Equation 12)

Here, β k and G ′ k are the k-th code vector in the two-dimensional gain codebook stored in the gain codebook 380. Further, δ (j) indicates the j-th shift amount, and g ′ k indicates the selected gain code vector. The index indicating the selected gain code vector and the code indicating the shift amount are assigned to the multiplexer 4.
Output to 00.

In modes 1 to 3, a code book for quantizing the amplitude of a plurality of pulses can be learned and stored in advance using an audio signal. Codebook learning methods are described, for example, by Linde et al., "An algorith
m for vector quantizationdesign, "(IEE
E Trans. Commun., Pp.84-95, January, 1980) (Reference 12).

The smoothing circuit 450 inputs the mode information,
When the mode information is a predetermined mode (for example, mode 0), at least one of the gain of the excitation signal in the gain code vector, the gain of the adaptive codebook, the RMS of the excitation signal, and the spectrum parameter is set in the time direction. Smoothing.

Here, the gain of the sound source signal is smoothed according to the following equation. Here, m indicates a subframe number.

(Equation 13)

The gain of the adaptive codebook is smoothed according to the following equation.

[Equation 14]

The smoothing of the RMS of the sound source signal follows the following equation.

(Equation 15)

The spectral parameters are smoothed according to the following equation. Here, a case where the smoothing is performed on the LSP is shown.

(Equation 16)

The weighting signal calculation circuit 360 receives the mode discrimination information and the smoothing signal of the smoothing circuit, and in the case of modes 1 to 3, obtains the driving sound source signal v (n) based on equation (17).

[Equation 17] v (n) is output to the adaptive codebook circuit 500.

In the case of the mode 0, the driving sound source signal v (n) is obtained based on the equation (18).

(Equation 18) v (n) is output to the adaptive codebook circuit 500.

Next, the spectrum parameter calculation circuit 200
Output parameter, spectrum parameter quantization circuit 21
Using the output parameter of 0 and the output parameter of the smoothing circuit 450, the response signal xw (n) is calculated for each subframe.
Here, in modes 1 to 3, calculation is performed by equation (19) and output to the response signal calculation circuit 240.

[Equation 19]

On the other hand, in mode 0, the LSP parameters smoothed in the smoothing circuit 450 are input, converted into smoothed linear prediction coefficients, and a response signal is calculated by the equation (20). Output to 240.

(Equation 20)

Next, a second embodiment of the present invention will be described with reference to FIG. The demultiplexer 500 extracts an index indicating a gain code vector from the received signal,
An index indicating the delay of the adaptive codebook, information of a sound source signal, an index of a sound source code vector, and an index of a spectrum parameter are input, and each parameter is separated and output.

Gain decoding circuit 510 receives the index of the gain code vector and the mode discrimination information, reads out the gain code vector from gain code book 380 according to the index, and outputs it.

The adaptive codebook circuit 520 receives the mode discrimination information and the delay of the adaptive codebook, generates an adaptive code vector, multiplies the gain of the adaptive codebook by the gain code vector, and outputs the product.

In the sound source signal restoring circuit 540, when the mode discrimination information is Modes 1 to 3, the sound source code book 351
A sound source signal is generated using the polarity code vector read from, the pulse position information and the gain code vector, and output to the adder 550. Mode 0 is mode 0
In the case of (1), a sound source signal is generated from the pulse position, the shift amount of the position, and the gain code vector, and output to the adder 550.

The adder 550 is connected to the adaptive codebook circuit 5
20 and the output of the sound source signal restoration circuit 540,
A driving sound source signal v (n) is generated, and the adaptive code book circuit 52
0 and output to the synthesis filter 560.

The spectrum parameter decoding circuit 570
The spectrum parameters are decoded, converted into linear prediction coefficients, and output to the synthesis filter circuit 560.

The synthesis filter circuit 560 generates the driving sound source signal.
v (n) and a linear prediction coefficient are input, and a reproduced signal s (n) is calculated.

The post-filter circuit 600 generates the reproduced signal s
(n) is subjected to post-filtering to mask quantization noise, and a post-filtering output signal S
Output _p (n). Here, the transfer characteristic of the post filter is expressed by equation (25).

(Equation 21)

The inverse post / synthesis filter circuit 610 forms an inverse filter of the post filter and the synthesis filter, and calculates a residual signal e (n). Here, the transfer characteristic of the inverse filter is expressed by equation (26).

(Equation 22)

The smoothing circuit 620 smoothes at least one of the gain of the excitation signal, the gain of the adaptive codebook, the RMS of the residual signal, and the spectral parameter in the gain code vector in the time direction. Here, the smoothing of the gain of the sound source signal, the smoothing of the gain of the adaptive codebook, and the smoothing of the spectral parameters are respectively performed by Equation (1)
Follow 7), (18) and (20). The RMS smoothing of the residual signal e (n) complies with Equation (27). Here, RMSe (m) indicates the RMS of the residual signal of the m-th subframe.

(Equation 23)

The smoothing circuit 620 restores the driving sound source signal using the smoothed parameters. Here, Expression (28) shows a case where the RMS of the residual signal is smoothed to restore the driving excitation signal.

(Equation 24)

The synthesis filter 560 outputs the driving sound source signal obtained using the smoothed parameters.

[Outside 1] To input the playback signal.

[Outside 2] calculate. In this case, a smoothed linear prediction coefficient may be used.

The post-filter 600 receives the reproduction signal, performs post-filtering, and performs a final reproduction signal after the post-filtering.

[Outside 3] Is output.

FIG. 3 is a block diagram showing a third embodiment. In FIG. 3, components denoted by the same reference numerals as those in FIG. 2 operate in the same manner, and a description thereof will be omitted.

In FIG. 3, an inverse post filter circuit 63
0, the smoothing circuit 640 receives discrimination information from the demultiplexer 500, and performs each operation when the discrimination information indicates a predetermined mode (for example, mode 0). Each operation is performed by the inverse post-synthesis filter circuit 610 of FIG.
Since each is the same as the smoothing circuit 620, the description is omitted.

[0078]

As described above, according to the speech coding apparatus of the present invention, at least one of the spectral parameter, the adaptive codebook gain, the excitation codebook gain, and the RMS of the excitation signal is smoothed in the time direction. Since the synthesized signal is locally reproduced using the converted signal, it is possible to suppress the local time variation of the parameter in the background noise section even at a low bit rate, even for voices with background noise superimposed, and There is an effect that coded speech with little deterioration can be provided.

According to the speech decoding apparatus of the present invention,
From the signal post-filtered on the decoding side, a residual signal is obtained by inverse post-synthesis filter processing.
RMS, received spectral parameters, adaptive codebook gain, at least one of the sound source codebook gains are smoothed in the time direction to re-synthesize the signal, post-filtered again and the final synthesized signal , So that the processing can be added as a complete post-processing without any modification to the conventional decoding device. There is an effect that a local temporal variation of a parameter can be suppressed, and a synthesized voice with less deterioration in sound quality can be provided.

Further, according to the speech decoding apparatus of the present invention, the parameter is smoothed in a predetermined mode or in a case where a feature value exists in a predetermined area. Can be performed only in the section (for example, a silent section), and the background noise portion can be satisfactorily encoded even if the speech on which the background noise is superimposed is encoded at a low bit rate without affecting the speech section. There is an effect that can be.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a first embodiment of a speech encoding apparatus according to the present invention.

FIG. 2 is a block diagram showing a second embodiment of the speech encoding apparatus according to the present invention.

FIG. 3 is a block diagram showing a third embodiment of the speech coding apparatus according to the present invention.

[Explanation of symbols]

 110 Frame division circuit 120 Subframe division circuit 200 Spectrum parameter calculation circuit 210 Spectrum parameter quantization circuit 211 LSP codebook 230 Perceptual weighting circuit 235 Subtraction circuit 240 Response signal calculation circuit 310 Impulse response calculation circuit 350, 355, 356, 357 Sound source quantum Circuit 351 sound source codebook 360 weighting signal calculation circuit 365,370 gain quantization circuit 380,515 gain codebook 400 multiplexer 450 smoothing circuit 500 adaptive codebook circuit 520 demultiplexer 510 gain decoding circuit 520 adaptive codebook circuit 540 sound source restoration circuit 550 Addition circuit 560 Synthesis filter circuit 570 Spectrum parameter decoding circuit 600 Post filter circuit 610, 630 Inverse post / synthesis filter circuit 620, 640 Smoothing circuit

[Procedure amendment]

[Submission date] January 23, 2001 (2001.1.2)
3)

[Procedure amendment 1]

[Document name to be amended] Statement

[Correction target item name] Claims

[Correction method] Change

[Correction contents]

[Claims]

[Procedure amendment 2]

[Document name to be amended] Statement

[Correction target item name] 0016

[Correction method] Change

[Correction contents]

[0016] (8) The mode discrimination, voice decrypt apparatus of (7) to be performed for each frame.

[Procedure amendment 3]

[Document name to be amended] Statement

[Correction target item name] 0017

[Correction method] Change

[Correction contents]

[0017] (9) the feature amount, the audio decrypt apparatus of (7) is a pitch prediction gain.

[Procedure amendment 4]

[Document name to be amended] Statement

[Correction target item name] 0018

[Correction method] Change

[Correction contents]

(10) In the mode discrimination, a pitch prediction gain obtained for each sub-frame is averaged over the entire frame,
Voice decrypt apparatus of (7) to be performed by comparing a plurality of predetermined threshold this value.

[Procedure amendment 5]

[Document name to be amended] Statement

[Correction target item name] 0019

[Correction method] Change

[Correction contents]

[0019] (11) the mode, unvoiced speech decrypted device transient period, weak voiced segment, in which corresponding substantially to the strong voiced segments (7).

[Procedure amendment 6]

[Document name to be amended] Statement

[Correction target item name] 0020

[Correction method] Change

[Correction contents]

(12) Locally reproducing a synthesized speech signal based on a signal obtained by smoothing at least one of a spectrum parameter, a gain of an adaptive codebook, a gain of a sound source codebook, and an RMS of a sound source signal in a time direction. voice decrypted device that.

[Procedure amendment 7]

[Document name to be amended] Statement

[Correction target item name] 0025

[Correction method] Change

[Correction contents]

The spectrum parameter quantization circuit 210 efficiently quantizes the LSP parameter of a predetermined sub-frame and outputs a quantization value for minimizing the following equation. Here, the spectrum parameter quantization circuit 210
Refer to the LSP codebook 211.

(Equation 1)

[Procedure amendment 8]

[Document name to be amended] Statement

[Correction target item name] 0034

[Correction method] Change

[Correction contents]

The impulse response calculation circuit 310 calculates the impulse response of the auditory weighting filter whose z-transform is expressed by the following equation.
hw (n) is calculated by a predetermined number L and output to the adaptive codebook circuit 470 and the sound source quantization circuit 350.

(Equation 4)

[Procedure amendment 9]

[Document name to be amended] Statement

[Correction target item name] 0035

[Correction method] Change

[Correction contents]

The mode discriminating circuit 300 uses the output signal of the frame dividing circuit to extract the characteristic amount and discriminate the mode for each frame. Here, as a feature, a pitch prediction gain can be used. The pitch prediction gain obtained for each subframe is averaged for the entire frame, this value is compared with a plurality of predetermined thresholds, and the mode is classified into a plurality of predetermined modes. Here, as an example, the type of the mode is 4. In this case, mode 0, 1, 2, 3
Respectively correspond approximately to an unvoiced section, a transient section, a weak voiced section, and a strong voiced section. The mode determination information is output to the sound source quantization circuit 350, the gain quantization circuit 365, and the multiplexer 400.

[Procedure amendment 10]

[Document name to be amended] Statement

[Correction target item name] 0036

[Correction method] Change

[Correction contents]

In the adaptive code book circuit 470 , the past excitation signal v (n) from the gain quantization circuit 370 is subtracted by the subtracter 2.
An output signal x ′ w (n) is input from 35 and an auditory weighting impulse response hw (n) is input from the impulse response calculation circuit 310. The delay T corresponding to the pitch is determined so as to minimize the distortion of the following equation, and the index representing the delay is calculated by the multiplexer 4.
Output to 00.

(Equation 5) In equation (8), the symbol * represents a convolution operation.

[Procedure amendment 11]

[Document name to be amended] Statement

[Correction target item name] 0039

[Correction method] Change

[Correction contents]

Further, the adaptive codebook circuit 470 performs pitch prediction according to the equation (10), and obtains a prediction residual signal ew (n).
Is output to the sound source quantization circuit 355.

(Equation 7)

[Procedure amendment 12]

[Document name to be amended] Statement

[Correction target item name] 0059

[Correction method] Change

[Correction contents]

The weighting signal calculation circuit 360 receives the mode discrimination information and the smoothing signal of the smoothing circuit, and obtains the driving sound source signal v (n) based on the equation ( 21 ) for modes 1 to 3.

[Equation 17] v (n) is output to the adaptive codebook circuit 470 .

[Procedure amendment 13]

[Document name to be amended] Statement

[Correction target item name] 0060

[Correction method] Change

[Correction contents]

In the case of the mode 0, the driving sound source signal v (n) is obtained based on the equation ( 22 ).

(Equation 18) v (n) is output to the adaptive codebook circuit 470 .

[Procedure amendment 14]

[Document name to be amended] Statement

[Correction target item name] 0061

[Correction method] Change

[Correction contents]

Next, the spectrum parameter calculation circuit 200
Output parameter, spectrum parameter quantization circuit 21
Using the output parameter of 0 and the output parameter of the smoothing circuit 450, the response signal xw (n) is calculated for each subframe.
Here, in modes 1 to 3, calculation is performed by equation ( 23 ), and the calculated signal is output to the response signal calculation circuit 240.

[Equation 19]

[Procedure amendment 15]

[Document name to be amended] Statement

[Correction target item name] 0062

[Correction method] Change

[Correction contents]

On the other hand, in mode 0, the LSP parameters smoothed by the smoothing circuit 450 are input, converted into smoothed linear prediction coefficients, and a response signal is calculated by the equation ( 24 ). Output to 240.

(Equation 20)

[Procedure amendment 16]

[Document name to be amended] Statement

[Correction target item name] Explanation of sign

[Correction method] Change

[Correction contents]

[Explanation of Signs] 110 Frame division circuit 120 Subframe division circuit 200 Spectrum parameter calculation circuit 210 Spectrum parameter quantization circuit 211 LSP codebook 230 Auditory weighting circuit 235 Subtraction circuit 240 Response signal calculation circuit300 Mode discrimination circuit  310 Impulse response calculation circuit 350, 355, 356, 357 Sound source quantization circuit 351 Sound source codebook 360 Weighted signal calculation circuit 365, 370 Gain quantization circuit 380, 515 Gain codebook 400 Multiplexer 450 Smoothing circuit470 Adaptive codebook circuit500 Demultiplexer 510 Gain decoding circuit 520 Adaptive codebook circuit 540 Sound source restoration circuit 550 Addition circuit 560 Synthesis filter circuit 570 Spectrum parameter decoding circuit 600 Post filter circuit 610, 630 Inverse post / synthesis filter circuit 620, 640 Smoothing circuit

[Procedure amendment 17]

[Document name to be amended] Drawing

[Correction target item name] Figure 3

[Correction method] Change

[Correction contents]

FIG. 3

Claims (14)

[Claims] A spectrum parameter calculator for inputting an audio signal and obtaining and quantizing a spectrum parameter for each predetermined frame; dividing the frame into a plurality of subframes; An adaptive codebook section for obtaining a delay and a gain from an encoded sound source signal using an adaptive codebook, predicting an audio signal and obtaining a residual, and quantizing and outputting the sound source signal of the audio signal using the spectrum parameter. In a speech encoding apparatus having a sound source quantization unit and a gain quantization unit for quantizing the gain of the adaptive codebook and the gain of the sound source signal, a predetermined feature amount is extracted from the speech signal and extracted. A mode discriminator for discriminating which mode among a plurality of predetermined modes based on the amount, When the output is in a predetermined mode, a smoothing circuit is provided for smoothing at least one of the gain of the excitation signal, the gain of the adaptive codebook, the spectrum parameter, and the level of the excitation signal in the time direction. The synthesized signal is locally reproduced using the smoothed signal, and the output of the spectrum parameter calculation unit, the output of the discrimination unit, the output of the adaptive codebook unit, the output of the sound source quantization unit, and the gain are output. A speech encoding device comprising: a multiplexer unit that combines and outputs an output of a quantization unit. 2. The speech encoding apparatus according to claim 1, wherein said mode discriminating circuit discriminates a mode for each frame. 3. The speech coding apparatus according to claim 1, wherein the feature quantity is a pitch prediction gain. 4. The mode discriminating circuit averages a pitch prediction gain obtained for each sub-frame for the entire frame, and compares this average with a plurality of predetermined thresholds to determine a plurality of predetermined modes. The speech encoding device according to claim 1, wherein the speech encoding device is classified as: 5. The speech coding apparatus according to claim 1, wherein the plurality of predetermined modes substantially correspond to an unvoiced section, a transient section, a weak voiced section, and a strong voiced section. 6. A spectrum parameter as voice information,
A demultiplexer circuit for inputting and separating a pitch, a gain and a sound source signal, and a sound source signal restoring circuit for restoring a sound source signal from the separated pitch, sound source signal and gain,
Speech decoding comprising a synthesis filter circuit for synthesizing a speech signal using the restored sound source signal and the spectrum parameter, and a post-filter circuit for inputting the synthesized speech and performing post-filtering using the spectrum parameter. In the apparatus, an inverse filter circuit that performs an inverse post-filtering and an inverse synthesis filtering from the output signal of the post-filter circuit and the spectral parameter to estimate a sound source signal,
A smoothing circuit for performing time-direction smoothing on at least one of the level of the estimated sound source signal, the gain, and the spectral parameter; and inputting the smoothed signal to the synthesis filter circuit. An audio decoding apparatus characterized in that the synthesized signal is input to the post-filter circuit to synthesize an audio signal. 7. A demultiplexer circuit for inputting and separating discrimination information indicating a mode determined based on a characteristic amount of a speech signal to be decoded, a spectrum parameter, a pitch, a gain, and a sound source signal, A sound source signal restoring circuit for restoring a sound source signal from a sound source signal and a gain, a synthesis filter circuit for synthesizing an audio signal using the restored sound source signal and the spectrum parameter, and inputting the synthesized speech to the spectrum parameter. And a post-filter circuit that performs post-filtering by using, when the discrimination information is in a predetermined mode, inverse post-filtering is performed based on the output signal of the post-filter circuit and the spectrum parameter. Estimate sound source signal by performing inverse synthesis filtering An inverse filter circuit, a smoothing circuit that performs time-direction smoothing on at least one of the level of the estimated sound source signal, the gain, and the spectrum parameter, and the smoothed signal to the synthesis filter circuit. An audio decoding apparatus, comprising: inputting an input synthesized signal to the post-filter circuit to synthesize an audio signal; 8. The speech encoding apparatus according to claim 7, wherein said mode discrimination is performed for each frame. 9. The speech coding apparatus according to claim 7, wherein said feature amount is a pitch prediction gain. 10. The method according to claim 1, wherein the mode discrimination is performed by averaging the pitch prediction gain obtained for each sub-frame in the entire frame, and comparing this value with a plurality of predetermined thresholds. Item 8. The speech encoding device according to Item 7. 11. The speech coding apparatus according to claim 7, wherein said mode substantially corresponds to an unvoiced section, a transient section, a weak voiced section, and a strong voiced section. 12. A local reproduction of a synthesized speech signal based on a signal obtained by smoothing at least one of a spectrum parameter, an adaptive codebook gain, a sound source codebook gain, and a sound source signal RMS of a sound signal in a time direction. A speech coding apparatus characterized by the above-mentioned. 13. A decoding apparatus for decoding a residual signal from a post-filtered signal on the decoding side of an audio signal by inverse post-synthesis filter processing, the RMS of the residual signal, a reception spectrum parameter, a gain of an adaptive codebook, and a sound source code. Speech decoding characterized by re-synthesizing an audio signal based on a signal obtained by smoothing at least one of the book gains in the time direction, further performing post-filtering processing, and outputting a final synthesized signal. apparatus. 14. A mode determined on the decoding side of an audio signal from a post-filtered signal by inverse post-synthesis filter processing based on a characteristic amount of an audio signal to be decoded, or In the case where the feature quantity is in a predetermined area, based on a signal obtained by smoothing at least one of the RMS of the residual signal, the received spectrum parameter, the gain of the adaptive codebook, and the gain of the excitation codebook in the time direction. An audio decoding apparatus for re-synthesizing the audio signal, and further performing post-filtering processing to output a final synthesized signal.