JP3144244B2 - Audio coding device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an audio encoding apparatus for encoding an audio signal at a low bit rate, particularly at a high quality of 4.8 kb / s or less.

[0002]

2. Description of the Related Art As a method of encoding a speech signal at a low bit rate of 4.8 kb / s, for example, M. Sch
roeder and B.R. "Cod by Atal
e-excited linear predictic
on: High quality speech a
t very low bit rates ”(Pro
c. ICASSP, pp. 937-940, 1985
), And "Improved speech quality" by Kleijn et al.
and efficient vector qua
ntization in SELP "(Proc. IC
ASSP, pp. 155-158, 1988) (Reference 2).
e Excited LPC Coding) is known. In this method, the transmitting side extracts linear parameters (LPC) analysis from the audio signal for each frame (for example, 20 ms) to extract spectral parameters representing the spectral characteristics of the audio signal, and further converts the frame into subframes (for example, 5 ms). , And the parameters in the adaptive codebook (delay parameters and gain parameters corresponding to the pitch period) based on the past excitation signals for each subframe
, The pitch of the audio signal of the sub-frame is predicted by an adaptive codebook, and the residual signal obtained by pitch prediction is applied to a sound source codebook (vector quantization code) composed of a predetermined type of noise signal. Book), the excitation signal is quantized by selecting the optimal excitation code vector and calculating the optimal gain. 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.

[0003]

In the above conventional example, when the bit rate is reduced to about 4.8 kb / s or less, the number of bits of the sound source codebook becomes insufficient, and particularly, the female sound of a short pitch or the child's The sound quality was sharply degraded for the voice.

As a method of improving sound quality for these problems, sound quality can be improved by performing search while weighting with a weighting filter using pitch when searching for a sound source codebook or an adaptive codebook. . This method is called harmonic waiting. The details of this method are described in "Techniques for improving" by Gerson et al.
the performances of CELP
−type speech coders ”(Reference 3). An outline of this method is described below. Here, for the sake of simplicity, a case where the order of the pitch weighting filter is the first order is described.

Hc (z) = 1−εβz− ^T (1) Here, β, T, and ε represent a pitch gain, a pitch period, or a delay and a weighting factor obtained by an adaptive codebook, respectively. However, 0 ≦ ε ≦ 1.

In the conventional method, when searching for the sound source codebook, the search is performed while passing the code vector through the filter of equation (1).

SUMMARY OF THE INVENTION An object of the present invention is to solve the above-mentioned problems and to provide a speech encoding apparatus having a relatively small amount of calculation and having good sound quality at 4.8 kb / s or less.

[0008]

According to a first aspect of the present invention, a speech signal is inputted, and a parameter calculating means for obtaining a spectrum parameter and a pitch from the speech signal is inputted. Impulse response calculation means for obtaining an impulse response;
A speech code, comprising: a code book including a plurality of types of code vectors; and a vector quantization unit that searches for at least one code vector using the impulse response when searching for the code vector. The apparatus is obtained.

According to the second invention, an audio signal is input,
Parameter calculation means for obtaining a spectrum parameter and a pitch from the voice signal, adaptive codebook means for obtaining a delay corresponding to the pitch, and an impulse response for obtaining the impulse response of the weighting filter by inputting the spectrum parameter and the pitch obtained by the parameter calculation means Calculating means for searching the adaptive codebook using either the impulse response or the weighting filter, having a codebook consisting of a plurality of types of code vectors, and at least one code when searching for the code vector. A speech coding apparatus characterized by having a vector quantization means for searching for a vector using the impulse response is obtained.

According to the third aspect, an audio signal is input,
A spectrum parameter calculation means for obtaining a spectrum parameter from the voice signal, an adaptive codebook means for obtaining a delay corresponding to a pitch using an adaptive codebook, and an impulse response for inputting the spectrum parameter and the delay and obtaining an impulse response of a weighting filter Calculating means, and a codebook comprising a plurality of types of code vectors, and having a vector quantization means for searching for at least one code vector using the impulse response when searching for the code vector. Is obtained.

[0011]

The operation of the speech coding apparatus according to the present invention will be described.

In the first invention, an audio signal is divided into frames (for example, 40 ms), and further divided into sub-frames (8 m
s). A spectrum parameter and a pitch period are extracted from the audio signal. A weighting filter based on a pitch spectrum having the following transfer characteristic is formed, and its impulse response h _cw (n) is calculated by a predetermined point L.

H _cw (z) = H _w (z) H _c (z) (2) where

[0014]

(Equation 1)

## EQU1 ## A _i is a linear prediction coefficient obtained from the spectrum parameter, and γ ₁ and γ ₂ are weighting coefficients for controlling weighting, respectively.

A vector quantization codebook for quantizing a speech signal or a sound source signal is previously provided for each subframe, and a predetermined number (2 ^B, where B is a bit of the vector quantization codebook) ) Are stored. A code vector is searched for at least one code vector c _j (n) using a distortion measure of the following equation.

[0017]

(Equation 2)

Here,

[0019]

(Equation 3)

## EQU1 ## Also, L ≦ N, and usually L <
Take N. N indicates the subframe length. However,

[0021]

(Equation 4)

Here, c _j (n) indicates the j-th code vector.

In the second invention, the search for the adaptive codebook uses the impulse response h _cw (n) of the pitch spectrum weighting filter calculated using the spectrum and the pitch obtained from the speech signal.

The operation of the adaptive code book will be briefly described. The adaptive codebook delays to minimize:
Calculate the gain.

[0025]

(Equation 5)

Here, β and T represent the gain and delay of the adaptive codebook, respectively. x _w (n) is a perceptual weighting signal.

Next, the sound source codebook is searched. At this time, at least one code vector is searched using the equations (4) to (7).

In the third aspect, the delay and gain of the adaptive codebook are searched so as to minimize the following equation.

[0029]

(Equation 6)

Here, h _w (n) is the impulse response of the weighting filter having the transfer characteristic of the above equation (3).

After obtaining the delay and gain by the adaptive codebook, the impulse response of the weighting filter based on the pitch spectrum is calculated by the equations (1) to (3) using the delay, gain and spectrum parameters. Thereafter, the sound source codebook is searched for in the same manner as in the first invention.

The operation of the present invention has been described above.

[0033]

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

In the figure, an audio signal is input from an input terminal 100, a frame dividing circuit 110 divides the audio signal for each frame (for example, 40 ms), and a sub-frame dividing circuit 120 divides the audio signal of the frame into a shorter frame. Divide into subframes (for example, 8 ms)

The spectrum parameter analysis circuit 200 cuts out the speech signal by applying a window (for example, 24 ms) longer than the subframe length to the speech signal of at least one subframe, and sets the spectrum parameter to a predetermined order (for example, (P = 10th order) is calculated. Since the spectral parameters change greatly with time, especially in the transitional interval between consonants and vowels, it is desirable to analyze every short time.However, this increases the amount of computation required for the analysis. , Any Q in the frame
Spectral parameters are calculated for a number of (Q> 1) subframes (eg, Q = 3, first, third and fifth subframes). Then, in the sub-frames not analyzed (here, the second and fourth sub-frames), the spectral parameters of the first and third sub-frames and the third and fifth sub-frames were linearly interpolated on the LSP described later. Are used as spectral parameters. Here, a well-known LPC analysis, Burg analysis, or the like can be used for calculating the spectrum parameters. Here, Burg analysis is used. For details of the Burg analysis, see the book entitled "Signal Analysis and System Identification" written by Nakamizo (Corona Publishing Co., 1988), 82-.
The description is omitted because it is described on page 87 (Reference 4).

Further, in the spectrum parameter analysis circuit, the linear prediction coefficient αi (i
= 1 to 10) are converted into LSP parameters suitable for quantization and interpolation. Here, the conversion from the linear prediction coefficient to the LSP is performed by a paper entitled "Speech Information Compression by Line Spectrum Pair (LSP) Speech Analysis / Synthesis Scheme" by Sugamura et al. −606, 19
1981) (Reference 5). That is,
The linear prediction coefficients obtained by the Burg method in the first, third, and fifth subframes are converted into LSP parameters, and the LSPs of the second and fourth subframes are obtained by linear interpolation.
The LSP of the subframe is inversely transformed back to the linear prediction coefficient, and the linear prediction coefficient α _i ^l (i =
1 to 10 and l = 1 to 5) are output to the audibility weighting circuit 230. Further, the LSP of the first to fifth subframes is output to spectrum parameter quantization circuit 210.

The spectrum parameter quantization circuit 210 efficiently quantizes LSP parameters of a predetermined subframe. In the following, it is assumed that vector quantization is used as the quantization method, and that the LSP parameter of the fifth subframe is quantized. LSP
A well-known method can be used for the vector quantization of the parameter. A specific method is described in, for example,
No. 71500 (Document 6) and JP-A-4-363000
Japanese Patent Application Laid-Open Publication No. 5-6199 (Reference 8), Nomura et al. "L
SP Coding Usage VQ-SVQ Wi
the Interpolation in 4.075
kbps M-LCELP Speech Code
r "(Proc. Mobile Mul
timedia Communications, p
p. B. 2.5, 1993) (Reference 9) and the like, and a description thereof is omitted here.

Further, the spectrum parameter quantization circuit 2
In 10, the LSP parameters of the first to fourth subframes are restored based on the LSP parameters quantized in the fifth subframe. Here, the LSP of the first to fourth subframes is restored by linearly interpolating the quantized LSP parameter of the fifth subframe of the current frame and the quantized LSP of the fifth subframe of one past 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, for each candidate, the cumulative distortion is evaluated, and a combination of the candidate for minimizing the cumulative distortion and the interpolation LSP is determined. Be able to choose. For details, for example, Japanese Patent Application No. 5-8737 (Reference 10) can be referred to.

The LSPs of the first to fourth sub-frames and the quantized LSP of the fifth sub-frame, which have been restored as described above, are calculated for each sub-frame by a linear prediction coefficient α ′ _i ^l (i = 1 to 10, l =
1 to 5), and outputs the result to the impulse response calculation circuit 310. In addition, an index representing the code vector of the quantized LSP of the fifth subframe is output to the multiplexer 400.

In the above, instead of linear interpolation, LS
The interpolation parameters of P are prepared for a predetermined number of bits (for example, 2 bits), and the LSPs of 1 to 4 subframes are restored for each of these patterns, and the code vector for minimizing the accumulated distortion and A set of interpolation patterns may be selected. By doing so, the transmission information increases by the number of bits of the interpolation pattern, but the temporal change in the LSP frame can be represented more precisely. Here, the interpolation pattern is LS for training.
It may be created by learning in advance using P data, or a predetermined pattern may be stored. As the predetermined pattern, for example, T.I. "Im" by Taniguchi et al.
provedCELP speech coding
at 4 kb / s and below "(Proc. ICSLP, pp. 41-44.199).
2) A pattern described in (Reference 11) or the like can be used.

In order to further improve the performance, after selecting an interpolation pattern, an error signal between the true value of the LSP and the interpolation value of the LSP is determined in a predetermined subframe, and the error signal is calculated. Further, it may be represented by an error codebook. For details, reference can be made to the aforementioned reference 9.

The perceptual weighting circuit 230 first receives the linear prediction coefficient α _i ^l (i = 1 to 10, l =
1 to 5), weighting the spectrum of the audio signal of the sub-frame by the formula (3),
The weighting signal x _w (n) is output once to the pitch extraction circuit.

The pitch extraction circuit 232 outputs the weighted signal x
_{From w} (n), the pitch period T,
Calculate the gain β.

[0044]

(Equation 7)

The response signal calculation circuit 240 receives the linear prediction coefficient α _i ^l for each subframe from the spectrum parameter calculation circuit 200, and quantizes, interpolates and restores the linear prediction coefficient α _i ^l from the spectrum parameter quantization circuit 210. The coefficient α ′ _i ^l is input for each sub-frame, and a response signal with the input signal d (n) = 0 for one sub-frame is calculated using the stored value of the filter memory, and the subtractor 235
Output to

The subtractor 235 subtracts the response signal for one subframe from x _w (n) and outputs x ′ _w (n) to the adaptive codebook circuit 300.

The impulse response calculation circuit 310 calculates the impulse response h _cw (n) of the weighting filter in which the z-transform is expressed by the above equation (1) and the impulse response h _w (n) in which the z-transform is expressed by the equation (1) ) Is calculated by a predetermined number, and h _w (n) is calculated by the adaptive codebook circuit 3
00, h _cw (n) is output to the sound source search circuit 350.

The adaptive codebook circuit 300 calculates a pitch parameter. An impulse corresponding to the obtained delay value is output to multiplexer 400. Further, pitch prediction is performed by the adaptive codebook according to the following equation, and an adaptive codebook prediction difference signal z (n) is output.

Z (n) = x ′ _w (n) −b (n) (11) where b (n) is an adaptive codebook output signal and can be expressed by the following equation.

B (n) = β · v (n−T) * h _w (n) (12) Here, β and T indicate a gain and a delay of the adaptive codebook, respectively. v (n) is an adaptive code vector.

The pitch weighting circuit 315 performs pitch weighting on b (n) according to the above equation (1), and outputs the result to the sound source search circuit 350.

The sound source search circuit 350 uses h _cw (n) for a part or all of the sound source code vectors stored in the sound source code book 351 to obtain (4) to (7).
The sound source code vector is searched according to the formula.

The gain quantization circuit 365 reads the gain code vector from the gain code book 366, and applies the excitation code vector and the gain code vector to the selected excitation code vector so as to minimize the expression (13). Select a combination.

[0054]

(Equation 8)

Here, β ′ _k and γ ′ _k are the k-th code vectors in the two-dimensional gain codebook stored in the gain codebook 366.

The index representing the selected sound source code vector and gain code vector is multiplexed by the multiplexer 400.
Output to

The weighting signal calculation circuit 360 inputs the output parameters of the spectrum parameter calculation circuit and the respective indexes, reads out the corresponding code vectors from the indexes, and obtains the drive excitation signal v (n).

Next, the spectrum parameter analysis circuit 20
Weighting signal sw using an output parameter of 0 and an output parameter of the spectrum parameter quantization circuit 210.
(N) is calculated for each subframe, and the response signal calculation circuit 2
Output to 40.

The description of the embodiment corresponding to the first invention has been completed.

FIG. 2 is a block diagram showing an embodiment of the second invention. In FIG. 2, components denoted by the same reference numerals as those in FIG. 1 have the same functions as those in FIG.

The pitch weighting circuit 510 performs pitch weighting on the weighted signal x _w (n) according to the above equation (1), and obtains the pitch / spectrum weighted signal x _w (n).
Output _cw (n).

The response signal calculation circuit 520 receives the linear prediction coefficient α _i ^l for each subframe from the spectrum parameter calculation circuit 200, and quantizes and interpolates and restores the linear prediction coefficient α _i ^l from the spectrum parameter quantization circuit 210. The coefficient α ′ _i ^l is input for each subframe, and T and β are input from the pitch extraction circuit 232 to form a filter represented by the transfer characteristic of the equation (2). Using the value, a response signal with the input signal d (n) = 0 is calculated for one subframe and output to the subtractor 535.

The subtractor 535 subtracts the response signal for one subframe from x _cw (n), and outputs x ′ _cw (n) to the adaptive codebook circuit 540.

The impulse response calculation circuit 550 calculates the impulse response h _cw (n) of the weighting filter whose z-transform is represented by the above equation (1) by a predetermined number of points. Search circuit 35
Output to 0.

The adaptive code book circuit 540 determines a pitch parameter. An impulse corresponding to the obtained delay value is output to multiplexer 400. Further, pitch prediction is performed by the adaptive codebook according to the following equation, and an adaptive codebook prediction difference signal z (n) is output.

Z (n) = x ′ _cw (n) −b ′ (n) (14) where b ′ (n) is an adaptive codebook output signal and can be expressed by the following equation.

B ′ (n) = β · v (n−T) * h _cw (n) (15) where β and T indicate the gain and delay of the adaptive codebook, respectively. v (n) is an adaptive code vector.

The gain quantization circuit 560 reads the gain code vector from the gain code book 366, and applies the excitation code vector and the gain code vector to the selected excitation code vector so as to minimize the expression (16). Select a combination.

[0069]

(Equation 9)

Here, β ′ _k and γ ′ _k are the k-th code vectors in the two-dimensional gain codebook stored in the gain codebook 366.

The impulse representing the selected sound source code vector and gain code vector is converted to a multiplexer 400.
Output to

FIG. 3 is a block diagram showing an embodiment of the third invention. In FIG. 3, components denoted by the same reference numerals as those in FIG. 1 have the same functions as those in FIG. 1, and a description thereof will be omitted.

The impulse response calculation circuit 630 calculates an impulse response h _w (n) whose z-transform is expressed by the equation (3) by a predetermined number of points, and calculates h _w (n) into the adaptive codebook circuit 300. And h _cw (n) to the sound source search circuit 350.

The pitch weighting circuit 620 receives T and β from the adaptive codebook circuit 300, and performs pitch weighting on the adaptive codebook output signal b (n) according to the above equation (1). Output to

The impulse response calculation circuit 630 receives T and β from the adaptive codebook circuit 300 and calculates an impulse response h _cw (n) whose z-transform is expressed by the equation (2) by a predetermined number of points. And a sound source search circuit 350
Output to

Various modifications other than the above-described embodiment are possible without impairing the intention of the present invention.

As the spectral parameters, other well-known parameters other than the LSP can be used.

When calculating a spectrum parameter in at least one subframe in a frame, the spectrum parameter calculation circuit measures a change in RMS or a change in power between a previous subframe and the current subframe, and calculates a change in these changes. May be calculated for a plurality of sub-frames with large.
In this way, the spectral parameter is always analyzed at the voice change point, and performance degradation can be prevented even if the number of subframes to be analyzed is reduced.

For quantizing the spectral parameters, known methods such as vector quantization, scalar quantization, and vector-scalar quantization can be used.

Other well-known distance measures can be used for selecting an interpolation pattern in the spectrum parameter quantization circuit.

The search for delay in the adaptive code book circuit can be performed by filtering without using an impulse response.

In the sound source quantization circuit, a case where the code book has one stage has been described, but a two-stage or multi-stage configuration is also possible.

Further, as a distance scale at the time of searching for a sound source codebook and learning, or another learning scale, a learning method can be used.

The gain codebook learns in advance a codebook having a size several times larger than the number of transmission bits in total, and assigns a partial area of the codebook as a use area for each predetermined mode. ,
When encoding, the used area can be switched according to the mode and used.

[0085]

As described above, according to the present invention, the impulse response is calculated using the spectral parameter and the pitch, and at least one of the adaptive codebook and the excitation codebook is searched, and at least one of the excitation code vectors is searched. On the other hand, since the code vector is searched using the impulse response, when the bit rate is low, it is possible to improve a female voice or a child's voice with a short pitch, and a reproduced voice with little deterioration can be obtained. There is an effect that it can be provided.

[Brief description of the drawings]

FIG. 1 is a block diagram of one embodiment of the first invention.

FIG. 2 is a block diagram of one embodiment of the second invention.

FIG. 3 is a block diagram of one embodiment of the third invention.

[Explanation of symbols]

 Reference Signs List 110 frame division circuit 120 subframe division circuit 200 spectrum parameter analysis circuit 210 spectrum parameter quantization circuit 211 LSP codebook 230 auditory sensation weighting circuit 232 pitch extraction circuit 235,535 subtraction circuit 240,520 response signal calculation circuit 300,540 adaptive codebook Circuit 310, 550, 630 Impulse response calculation circuit 315, 510, 620 Pitch weighting circuit 350 Sound source search circuit 351 Sound source codebook 365, 560 Gain quantization circuit 366 Gain codebook 360, 530 Weighting signal calculation circuit 400 Multiplexer

Claims (1)

(57) [Claims] An audio signal is input and a spectrum is input from the audio signal.
Calculator of spectral parameters to obtain vector parameters
A stage (200, 210, 211); the audio signal and the past excitation signal;Impulse response andFrom
A delay is obtained, and the voice signal is pitch-predicted using the delay.
Adaptive codebook signal to subtract the adaptive codebook signal
Input means (235, 300), and the spectrum parameters and the delay are inputted and predetermined.
Represents the characteristics of a weighted filter with specified transfer characteristics
The impulse responseMeans for calculating impulse response
(630) and a plurality of codes for quantizing the output signal of the adaptive codebook.
Sound source code block consisting of several types of code vectors
(351), calculate the distortion between the synthesized signal and the audio signal, and calculate the distortion.
Sound source search means for searching for the code vector to be minimized
(350),The spectrum parameter, the delay, and the sound source
An index indicating the code vector searched by the search means.
Multiplexed with audio and output as encoded audio signal
Multiplexer (400) Speech coding device with
PlaceAnd  When calculating the distortion, the sound source searching means may select the sound
A predetermined number of code vectors in the source codebook
It is characterized in that the impulse response is used for torque.
Speech encoding device.