JP3252285B2 - Audio band signal encoding method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention encodes a voice band signal, particularly a voice signal and a voice band modem signal, with high quality at a low bit rate, especially about 8 to 4.8 kb / s. And a voice band signal encoding method for the same.

[Conventional technology]

As a method of encoding an audio signal at a low bit rate of about 8 to 4.8 kb / s, for example, M. Schroeder and B. At
al “Code-excited linear prediction: High Q
uality Speech at very low bit rates "(Proc.ICASSP,
pp, 937-940, 1985), a CELP (Code Excited LPC Coding) described in a paper (Reference 1) and the like is known. In this method, the transmitting side extracts a spectrum parameter representing a spectrum characteristic of a voice signal from a voice signal for each frame (for example, 20 ms), further divides the frame into small-section subframes (for example, 5 ms), and A pitch parameter representing a long-term correlation (pitch correlation) is extracted based on a past sound source signal, and a long-term prediction of a subframe speech signal is performed based on the pitch parameter. One kind of noise signal is selected so as to minimize the error power between a signal synthesized from a code book composed of a predetermined kind of noise signal and a speech signal, and an optimum gain is calculated. Then, an index and a gain representing the type of the selected noise signal, and a spectrum parameter and a pitch parameter are transmitted.

[Problems to be solved by the invention]

In the above-mentioned conventional method of Reference 1, in order to obtain high sound quality, it is generally necessary to extremely increase the bit size of a codebook composed of noise signals to 10 bits or more. There is a problem that an enormous amount of calculation is required to obtain a suitable noise signal (codeword). Furthermore, since the codebook is basically composed of a noise signal, there is a problem that the sound quality of the reproduced voice reproduced by the sound source signal selected from the codebook is accompanied by noise.

Further, in the conventional method, when a modem signal or the like in a voice band is input, the codebook is not optimal for the modem, so that there is a problem that characteristics are deteriorated.

SUMMARY OF THE INVENTION An object of the present invention is to solve the above-mentioned problems, and achieve a good sound quality at about 8 to 4.8 kb / s and a good characteristic for a modem signal in a voice band with a relatively small amount of computation and memory. Is to provide a voice band signal encoding method that can be obtained.

[Means for solving the problem]

A first invention divides an input discrete voice band signal into frames of a predetermined time length, obtains and outputs spectral parameters representing a spectral envelope of the input signal, and outputs the frames in a predetermined manner. Divided into small sections of time length, a signal reproduced based on a past sound source signal is obtained as a pitch parameter so as to be close to the input signal, a pitch prediction is performed on the input signal to obtain a residual signal, A distortion from the residual signal or the audio signal is reduced from a first codebook in which a plurality of types of code vectors previously learned and configured based on an audio signal or an audio band signal other than the audio signal are stored. A code vector is selected, and the difference signal is obtained from a second code book in which a code vector having a known statistical characteristic or a code vector configured by learning in advance is stored. A code vector for reducing distortion is selected, and a sound source of the input signal is represented and output using a code vector selected from the first code book and a code vector selected from the second code book. It is characterized by the following.

A second invention divides an input discrete voice band signal into frames of a predetermined time length, obtains and outputs a spectrum parameter representing a spectrum envelope of the input signal, and outputs the frame in a predetermined manner. Divided into small sections of time length, a signal reproduced based on a past sound source signal is obtained as a pitch parameter so as to be close to the input signal, a pitch prediction is performed on the input signal to obtain a residual signal, A distortion from the residual signal or the audio signal is reduced from a first codebook in which a plurality of types of code vectors previously learned and configured based on an audio signal or an audio band signal other than the audio signal are stored. Selecting a code vector and storing said residual signal from a second code book in which a code vector having known statistical characteristics or a code vector previously learned and stored is stored. Alternatively, a code vector for reducing distortion with the audio signal is selected, and at least one of the pitch parameter or the gain of the sound source signal is quantized and output using a third codebook configured in advance. And a weighted linear combination of the code vector selected from the code book, the code vector selected from the second code book, and the gain, and represents and outputs the excitation signal of the input signal.

[Action]

The operation of the voice band signal encoding method according to the present invention will be described.

In the first invention, a sound source signal representing an input voice band signal is obtained for each subframe obtained by dividing a frame so as to minimize the following expression.

Here, x (n) is a voice band signal, β, m are pitch parameters of pitch prediction (adaptive codebook) based on long-term correlation, that is, gain and delay, and v (n) is a past sound source signal. h (n) is the impulse response of the synthesis filter composed of spectral parameters, w
(N) shows the impulse response of the auditory weighting filter. The symbol * indicates a convolution operation. For details of w (n), reference can be made to Reference 1.

Also, d (n) indicates a sound source signal represented by a codebook, and as shown in the following equation, a codeword c ₁ (n) selected from the first codebook and a codeword c ₁ (n) selected from the second codebook. It is represented by a linear combination with the code word c ₂ (n).

Where γ ₁ and γ ₂ are the selected codewords c _1j (n),
Indicates the gain of c _2j (n). Therefore, in the present invention, since the sound source signal is represented by being decomposed into two types of codebooks, each codebook may be half the number of bits of the entire codebook. For example, if the number of bits in the entire codebook is 1
If the number of bits is 0, the first and second codebooks need only be 5 bits each, and the amount of calculation for codebook search can be greatly reduced.

When a noise codebook as described in Reference 1 is used as each codebook and the codebook is divided as shown in Expression (2), the codebook is characteristically degraded compared to a 10-bit codebook and has a total of 7 to 8 bits.
Only a bit worth of performance can be obtained.

Therefore, in the present invention, in order to obtain high performance, the first codebook is configured by learning in advance using training data. Here, the first codebook includes a codebook learned for an audio signal and a codebook learned for an audio band signal other than audio. In the following, a codebook optimally learned for a voice signal and a codebook optimally learned for a modem signal in a voice band are used as input signals for a voice signal and a modem signal in a voice band. It is assumed that the type is provided as a first codebook.

As a method of constructing a codebook by learning, for example,
“An Algorithm for Vector Quantizatio by Linde et al.
n Design "(IEEE Trans.COM-28, pp.84-9
5, 1980) (Reference 2).

As a distance measure at the time of learning, a square distance (Euclidean distance) is usually used, but in the present invention, an auditory weighting distance scale that has better performance than the square distance and is represented by the following equation is used.

Where t _j (n) is the j-th training data, c
₁ (n) is the code word of cluster 1. The centroid (representative codeword) of cluster 1 is cluster 1
Equation (4) or Equation (5) is determined to be minimized using the training data in.

In the equation (5), g indicates an optimum gain.

On the other hand, the second codebook is composed of a noise signal or random number signal whose statistical characteristics have been determined in advance, such as the Gaussian noise signal of Document 1, in order to relieve the training data dependency due to the first codebook. Use codebooks or codebooks with other characteristics. In addition,
By selecting the noise codebook based on a certain distance measure, the characteristics are further improved. See T. for details.
“Transform Coding of Speechusing by Moriya et al.
a Weighted Vector Quantizer, "(IEEE, J.
Sel, Areas, Commun., Pp. 425-431, 1989) (Reference 3). Further, a second signal is obtained by using an error signal between the signal reproduced by the first codebook and the input signal.
Can be constructed by learning.

In the second aspect of the invention, first, after selecting the optimum codeword from the second codebook, the pitch parameters gain and first, gain gamma ₁ of the second codebook, at least one gamma _2, Vector quantization is efficiently performed using a codebook (third codebook) configured in advance by learning.

For example, assuming that the gains of the first and second codebooks are vector-quantized, the search for the optimal codeword in the vector quantization selects the codeword that minimizes the following equation.

Here, γ ′ _i is the value of the vector-quantized gain indicated by each codeword. C _i (n) is a codeword selected from the first and second codebooks. (6)
In the formula, Then, the following equation is obtained from the equations (6) and (7).

here It is. Further, since the first term of the equation (8) is a constant, the search for the codeword may be selected to maximize the second term of the equation (8).

Further, in order to greatly reduce the calculation amount of the codebook search, a codeword may be selected according to the following equation.

The codebook for vector quantization of gain is configured by learning in advance using training data including a large amount of gain value. Reference 2
Can be used. Here, the distance scale at the time of learning is usually a square distance, but in order to further improve the characteristics, a distance scale represented by the following equation is used.

Here, γ _ti is gain data for training.
γ ′ _i1 is a representative codeword in cluster 1 of the gain codebook.

When the distance scale of the equation (10) is used, the centroid Sc _1i in the cluster 1 is obtained so as to minimize the following equation.

On the other hand, in order to greatly reduce the amount of calculation in learning, a distance scale expressed by the following equation using a normal square distance may be used.

[Embodiment] FIG. 1 is a block diagram showing an audio band signal encoding apparatus applied to an embodiment of an audio band encoding system according to the first invention.

In the figure, on the transmission side, an audio signal is input from an input terminal 100, and an audio signal for one frame (for example, 20 ms) is stored in a buffer memory 110.

The LPC analysis circuit 130 performs a well-known LPC analysis on the speech signal of the frame as a parameter representing the spectral characteristic of the speech signal of the frame, and calculates a predetermined order L. Reference 1 can be referred to for the specific calculation method. Next, LSP quantization circuit 1
40 quantizes the LSP parameter with a predetermined number of quantization bits, outputs the obtained code 1 _k to the multiplexer 260, decodes this, and further decodes the linear prediction coefficient a _i ′
(I = 1 to L) and output to the weighting circuit 200, the impulse response calculation circuit 170, and the synthesis filter 281. The method of encoding LSP parameters and converting LSP parameters to linear prediction coefficients is described in “Quantize D by Sugamura et al.
isign in LSP Speech Analysis-Synthesis "(IEEE J. Sel, Areas, Commun., pp.432-440, 1988)
(Reference 4) can be referred to.

The subframe division circuit 150 divides an input audio band signal of a frame into subframes. Here, for example, the frame length is 20 ms, and the subframe length is 5 ms.

The subtractor 190 subtracts the output of the synthesis filter 281 from the input signal divided into subframes and outputs the result.

The weighting circuit 200 performs well-known auditory weighting on the subtracted signal. Reference 1 can be referred to for details of the auditory sensation weighting function.

The adaptive codebook 210 inputs the input signal v (n) of the synthesis filter 281 via the delay circuit 206, and further outputs the weighted impulse response h from the impulse response output circuit 170.
_w (n), a weighting signal is input from the weighting circuit 200,
Pitch prediction is performed based on the long-term correlation, and delay M and gain β are calculated as pitch parameters. In the following description, the prediction order of the adaptive codebook 210 is set to 1, but may be set to a second or higher order. The calculation method of the delay M and the gain β in the first-order adaptive codebook is described in Kleijin “Improve
d speech quality and efficient vector quantization
in SELP "(ICASSP, pp. 155-158, 1988
Year) (Reference 5). Further, the obtained gain β is quantized and decoded by the gain quantizer 220 with a predetermined number of quantization bits to obtain a gain β ′, which is used to calculate a prediction signal _w (n) by the following equation, and 20
Output to 5. The delay M is output to the multiplexer 260.

In the above equation, v (n−M) is a past sound source signal, which is an input signal of the synthesis filter 281. h _w (n) is a weighted impulse response obtained by the impulse response calculation circuit 170.

The delay circuit 206 sets the synthesis filter input signal v (n) to 1
Output to adaptive codebook 210 after delaying by a subframe.

The subtractor 205 subtracts the output of the adaptive codebook 210 from the output signal of the weighting circuit 200, and calculates a residual signal e _w (n) of the following equation.
To the first codebook search circuit 230.

The impulse response calculation circuit 170 calculates the impulse response h _w (n) of the synthesis filter weighted by the auditory sense for a predetermined number L of samples. For a specific calculation method, reference can be made to Document 1 and the like.

The first codebook search circuit 230 searches for both the codebook 235 constructed by learning the speech signal and the codebook 236 constructed by learning the speech band modem signal, and uses the optimum Search for codeword c _1j (n). As mentioned in the section on operations,
235 and 236 are learned in advance using a training signal.

FIG. 2 is a block diagram showing a configuration of the first codebook search circuit 230. The search for the codeword follows the formula below.

(21) to determine the gamma ₁ that minimizes the expression (21)
Using the formula obtained by placing a 0 as partial differential equation with gamma _1.

However At this time, equation (21) is Becomes Here, since the first term of the equation (25) is a constant, all of the codebooks 235 and 236 are searched to maximize the second term, and a codeword c _j (n) is selected. Γ ₁ is calculated from

In FIG. 2, the cross-correlation sensation calculation circuit 410 includes (2
3) Calculate the equation, and the autocorrelation function calculation circuit 420 calculates (24)
The formula is calculated, and the determination circuit 430 calculates the formula (25), selects the code word c _j (n), and outputs an index representing the code word c _j (n).

Further, the following method can be used to reduce the amount of calculation required for searching the codebook.

However Here, μ (i) and v _k (i) indicate the autocorrelation of the i-th delay of h _w (n) and the autocorrelation of the i-th delay of the codeword c _1k (n), respectively.

The index indicating the codeword obtained by the above method is output to the multiplexer 260. Further, it outputs the selected codeword c _j (n) to multiplier 241. Also,
A determination code indicating which codebook to use from the two types of codebooks 235 and 236 is output to the multiplexer 260. That is, when the optimal codeword c _j (n) is selected from the codebook 235, a code indicating the codebook 235 is output, and when selected from the codebook 236, a code indicating the codebook 236 is output.

Multiplier 241, a gain gamma ₁ codeword c _j (n) is multiplied by the following expression to output to the synthesis filter 250 obtains the equation of the sound source signal q (n).

The synthesis filter 250 receives the output q (n) of the multiplier 241 and calculates and outputs a weighted synthesized signal y _w (n) according to the following equation.

The subtractor 255 subtracts y _w (n) from e _w (n) to obtain a second
To the codebook search circuit 270.

The second codebook search circuit 270 calculates an optimal codeword from the second codebook 275. The configuration of the second codebook search circuit can be basically the same as the configuration of the first codebook search circuit shown in FIG. Further, the same method as the first codebook search can be used as a codeword search method, and the second codebook can be constructed as described in the section of the operation, In order to relieve the training data dependency while maintaining high efficiency, a codebook composed of a random number sequence is used. Reference 2 can be referred to for a method of constructing a codebook composed of a random number sequence.

In addition, in order to reduce the amount of calculation for codebook search,
As the second codebook, an overlap random codebook can be used. For the construction method of the superimposed random number codebook and the codeword search method, reference can be made to the aforementioned reference 2.

The gain quantizer 286 is provided with a gain codebook 287 that is optimally created for a voice signal using the above-described equations (12) and (13) by learning in advance, and a gain quantizer 286 for a voice-band modem signal. And optimally created codebook 28
8 and vector quantize the gains γ ₁ and γ ₂ .

The above-mentioned equation (8) is used for selecting an optimal codeword at the time of vector quantization.

FIG. 3 is a block diagram showing the configuration of the gain quantizer 286. In the figure, the reproduction circuit 505 includes c ₁ (n), c
₂ (n) and h _w (n) are input, and s _w1 (n) and s _w2 (n) are obtained based on the above equations (9) and (10).

Cross-correlation function calculation circuit 500, auto-correlation function calculation circuit 510
_{Is, e w (n), s} w1 is the output of the reproducing circuit 505 (n), s _n2
(N) Input the codeword output from the gain codebooks 287 and 288, and calculate the second and subsequent terms of the above equation (8). The maximum value discriminating circuit 520 is a codebook 287,288
In addition, the maximum value of the second term or less in equation (8) is determined, and an index indicating the codeword at that time is output.

The gain decoding circuit 530 decodes the gain using the index and outputs the result. Then, the index of the codebook is output to multiplexer 260. The decoded values γ ₁ ′ and γ ₂ ′ of the gain are output to the multiplier 242.

The multiplier 242 multiplies the codewords c ₁ (n) and c ₂ (n) selected by the first and second codebooks by the gains γ ₁ ′ and γ ₂ ′ that are quantized and decoded, respectively. Output to the synthesis filter 281.

The synthesis filter 281 receives the output v (n) of the multiplier 242, obtains a synthesized voice for one frame by the following equation, and inputs a sequence of 0 for another frame to the filter to obtain a response signal sequence. One frame of the response signal sequence is subtracted by 19
Output to 0.

However The multiplexer 260 combines and outputs the output code sequences of the LSP quantizer 140, the first codebook search circuit 230, the second codebook search circuit 270, and the gain quantizer 286.

 This concludes the description of the first embodiment of the present invention.

FIG. 4 is a block diagram showing one embodiment of the second invention. In the figure, components having the same numbers as in FIG. 1 perform the same operations as in FIG.

In the figure, a quantizer 225 vector-quantizes the gain of an adaptive codebook using a codebook 226 that has been learned and configured in advance based on the above equations (16) and (17). The above equation (15) is used for selecting an optimal codeword. Further, the quantizer 225 outputs a code indicating the index of the selected codeword to the multiplexer 260, and quantizes and decodes the gain to output.

 This concludes the description of the embodiment of the second invention.

In the above description, the adaptive codebook gain and the gains of the first and second codebooks are not simultaneously optimized, but the gains of the adaptive codebook, the first codebook, and the second codebook are not described. , Perform simultaneous optimization and further improve the characteristics. As described in the section of the operation, this synchronization optimization is further improved when it is applied when obtaining the codewords of the first and second codebooks.

For example, after determining the delay and gain β of the adaptive codebook, the codewords c _1j (n),
When searching a gain gamma _1, for each codeword, simultaneously optimizing the β and gamma ₁ solving to minimize following equation.

from now on, here, It is.

Next, when determining the second codeword, the gain of the adaptive codebook, the first, second,
Optimize codebook gains simultaneously.

In order to reduce the amount of subtraction, the gain is optimized by the equation (33) at the time of searching the codeword of the first codebook, and is not performed at the time of searching the second codebook. Can also.

Further, in order to further reduce the amount of computation, when searching for a codeword in the codebook, the gain is not optimized, and when the codeword in the first codebook is selected, the adaptive codebook and the The gain of the first codebook is jointly optimized, and when the codeword of the second codebook is selected, the adaptive codebook and the first,
A configuration that simultaneously optimizes the gain of the second codebook may be used.

Further, in order to further reduce the amount of computation, after the code words of the first and second codebooks is selected, the gain beta ₁ of the adaptive codebook, first, the gain of the second codebook γ ₁ may be configured such that at the same time optimize three of gamma _2.

In addition, the selection of the optimal codeword in the vector quantization of the gain of the adaptive codebook and the gains γ ₁ and γ ₂ of the _first and _second codebooks is performed by using the above equations (18), Equation (11) can also be used.

In addition, the first codebook search circuit determines which of the codebooks 235 and 236 has been selected the best codeword for each subframe, and uses this determination signal when performing vector quantization of gain. Alternatively, the codebooks 287 and 288 may be selected in advance, and only one gain codebook may be searched. With such a configuration, the gain codebook does not need to search both, so that the amount of calculation can be reduced.

In the above-described embodiment, the first codebook search method may use other well-known methods other than the method of the embodiment. For example, the method described in Document 1 or the orthogonal transformation c ₁ of each codeword c _1j (n) in the codebook in advance.
(K) is obtained and stored, and for each sub-frame, the orthogonal transform H _w (k) of the weighted impulse response h _w (n) and the orthogonal transform E _w (k) of the residual signal e _w (n) May be obtained by a predetermined score, and the following expression may be used instead of the expressions (13) and (14).

Then, the equations (37) and (38) are inversely orthogonally transformed, and the cross-correlation function G _j and the auto-correlation function C _j are calculated.
A configuration in which a code word is searched and a gain is calculated according to the formula may be adopted. At this time, a Fourier transform, an FFT, a cosine transform, or the like can be used as the orthogonal transform. According to this method, the above (13), (14)
Since the convolution operation of the expression can be reduced to multiplication on the frequency axis, the amount of operation can be reduced.

As the second codebook search method, in addition to the method of the above-described embodiment, the method described above, the method described in the above-mentioned document 5, and other well-known good methods can be used.

As a method of constructing the second codebook, in addition to the method described in the above embodiment, for example, an enormous random number sequence is prepared in advance as a codebook, and the random number sequence is searched for training data by using them. And register it as a codeword from the one that is selected
Can be configured. This configuration method can also be applied to the configuration of the first codebook.

In the above embodiment, the gains of the first and second codebooks are vector-quantized. However, the gain β of the adaptive codebook and the gains γ ₁ and γ ₂ of the _first and _second codebooks are separately calculated. Alternatively, a configuration in which these are collectively vector-quantized can be adopted. At this time, a codebook optimally learning these codebooks for voice signals and a codebook optimally learning for voiceband modem signals are prepared. The creation of the codebook of the vector quantizer by learning at this time can be referred to the above-mentioned reference 2.

In the above embodiment, the order of the adaptive codebook is 1
However, a higher order than the second order can be used. Also,
The delay may be a decimal value instead of an integer value while keeping the order. For more information on these, see Marque
"Pitch Prediction with Fractional Delay
s in CELP Coding "(EUROSPEECH, pp.509-
513, 1989) (Reference 6). Although the characteristics are improved as described above, the amount of information necessary for transmitting the gain or the delay slightly increases.

Further, in the above embodiment, the K parameter and the LSP parameter are encoded as the spectrum parameters, and the LPC analysis is used as the analysis method, but other known parameters such as the LPC cepstrum,
Cepstrum, improved cepstrum, generalized cepstrum, mel cepstrum and the like can also be used. In addition, an optimal analysis method can be used for each parameter.

Further, the LPC coefficient obtained in the frame may be interpolated for each subframe on the LSP or linear prediction coefficient, and the adaptive codebook and the first and second codebooks may be searched using the interpolated coefficient. . With such a configuration, the sound quality is further improved.

In addition, the LSP coefficient is obtained by vector quantization or vector-scalar quantization by a known method,
It is possible to encode more efficiently. For the method of vector-scalar quantization, for example, the above-mentioned Document 3 can be referred to. In such a configuration, as a codebook for vector quantization, a codebook optimally learned for an audio signal and a codebook optimally learned for an audio band modem signal are provided. By selecting the best codeword from these, the characteristics can be further improved.

Further, in order to reduce the amount of calculation, the transmission side may omit the calculation of the influence signal. This eliminates the necessity of the synthesis filter 281 and the subtractor 190 on the transmission side, thus making it possible to reduce the amount of computation, but slightly lowers the sound quality.

In addition, in order to reduce the amount of calculation, the weighting circuit 200 may be arranged before the subframe division circuit 150, and the synthesis filter 281 may calculate the weighted synthesized signal by the following equation.

Here, γ is a weighting coefficient that determines the degree of hearing weighting.

Also, on the receiving side, an adaptive post filter that operates on at least one of the pitch and the spectral envelope may be added in order to make the quantization noise shaped to make it easier to hear. For the configuration of the adaptive post filter, see, for example, Kroon et al.
is-by-synthesis Predictive Coders for High Quali
ty Speech Coding at Rates between 4.8 and 16kb / s, "
(IEEE JSAC, vol. 6, 2, 353-363, 1988) (Reference 7).

As is well known in the field of digital signal processing, the autocorrelation function is represented by a power spectrum on the frequency axis,
Since the cross-correlation function corresponds to the cross-power spectrum, it can be calculated from them. These calculations are described in “Digital Signal Pr
ocessing "(Prentice-Hall, 1975) (Reference 8).

〔The invention's effect〕

As described above, according to the present invention, the codebook representing the sound source signal is separated into two stages, and the first codebook is obtained by optimally learning the audio signal in advance, Having a codebook optimally learned and configured for voice band signals other than voice signals, for example, data modem signals, selecting the best codeword from both,
The second codebook uses a codebook having predetermined statistical characteristics, and at least one of a gain of the codebook and a gain of an adaptive codebook based on pitch prediction is applied to a large amount of training signals in advance. By performing vector quantization using a learned and configured codebook, the amount of computation is smaller than in the conventional method, and even for audio signals, audio band signals other than audio signals, such as data modem signals, etc. Also, there is an effect that better characteristics can be obtained as compared with the related art. Further, the characteristics are further improved by optimizing the gain of the codebook.

Therefore, according to the present invention, 8 to 4.8 k
At a bit rate of b / s, there is a great effect that a good voice band signal coding scheme can be obtained.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing a configuration of an audio band signal encoding apparatus to which an embodiment of an audio band signal encoding system according to the first invention is applied, and FIG. 2 is FIG. FIG. 3 is a block diagram showing a configuration of the gain quantizer of FIG. 1, and FIG. 4 is an embodiment of a voice band signal encoding method according to the second invention. 1 is a block diagram illustrating a configuration of a speech band signal encoding device to which is applied. 110 buffer memory 130 LPC analysis circuit 140 LSP quantization circuit 150 subframe division circuit 170 impulse response calculation circuit 190, 205, 255 subtractor 200 weighting circuit 206 delay circuit 210 Adaptive codebook 220 Gain quantizer 225 Quantizer 286 Gain quantizer 230 First codebook search circuit 235,236 Codebook 250,281 Synthesis filter 241,242 Multiplier 270 ... Second codebook search circuit 275... Second codebook 287... Gain codebook 410,500... Cross-correlation function calculation circuit 420,510...

Continuation of the front page (58) Field surveyed (Int. Cl. ⁷ , DB name) G10L 19/00-19/14 H03M 7/30 H04B 14/04

Claims (2)

(57) [Claims] 1. An input discrete voice band signal is divided into frames of a predetermined time length, a spectrum parameter representing a spectrum envelope of the input signal is obtained and output, and the frame is output for a predetermined time. Divided into long sections, based on a past sound source signal, a pitch parameter is calculated so that a reproduced signal is close to the input signal, pitch prediction is performed on the input signal to obtain a residual signal, and a speech signal is obtained. Alternatively, a code vector that reduces distortion with the residual signal is selected from a first code book in which a plurality of types of code vectors configured by learning in advance based on audio band signals other than the audio signal are stored, A first sound source signal is generated using a code vector selected from the first code book, and a synthesized residual signal synthesized based on the first sound source signal and the first sound source signal are generated. A difference signal from the difference signal is generated, and a code vector having a known statistical characteristic or a code vector that reduces distortion from the difference signal is selected from a second code book in which a code vector configured by learning in advance is stored. And using a code vector selected from the first codebook and a code vector selected from the second codebook to represent and output a sound source of the input signal. Encoding method. 2. An input discrete voice band signal is divided into frames of a predetermined time length, a spectrum parameter representing a spectrum envelope of the input signal is obtained and output, and the frame is output for a predetermined time. Divided into long sections, based on a past sound source signal, a pitch parameter is determined so that a reproduced signal is close to the input signal, pitch prediction is performed on the input signal to determine a residual signal, and a speech signal is obtained. Alternatively, a code vector for reducing distortion with the residual signal or the audio signal from a first code book in which a plurality of types of code vectors previously learned and configured based on audio band signals other than the audio signal are stored. And generates a first sound source signal using the code vector selected from the first codebook, and synthesizes the first sound source signal based on the first sound source signal. Generating a difference signal between the resultant residual signal and the residual signal; and distorting the differential signal from the second codebook having a known statistical characteristic or a code vector stored by learning in advance. Is selected, and at least one of the pitch parameter or the gain of the excitation signal is quantized and output using a third codebook configured in advance, and the code selected from the first codebook is output. A speech band signal encoding method characterized by expressing and outputting a sound source of the input signal by weighted linear combination of a vector, a code vector selected from the second codebook, and the gain.