JP3212123B2 - Audio coding device - Google Patents

Abstract

PURPOSE:To reduce deterioration in the quality of a speech even if a parameter in a 2nd noise code book has an error by not using the 2nd noise code book having low priority to update an adaptive code book. CONSTITUTION:Noise parameters stored in a 1st noise code book 120 are set higher in priority than noise parameters stored in the 2nd noise code book 150. The indexes of the adaptive code book 110, the code of a gain outputted from a gain encoding circuit 140, and the index and gains of the noise code book 120 are protected with error correction codes having high correcting ability. Only the indexes and gains of the adaptive code book 110 and noise code book 120 are used to generate driving codes to be stored in the adaptive code book 110 sequentially.

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, more particularly, to a speech coding apparatus suitable for coding a speech signal at a low bit rate of about 8 kbps or less.

[0002]

2. Description of the Related Art A technology for encoding a speech signal at a low bit rate and a high efficiency is an important technology for effective use of radio waves and reduction of communication costs in mobile communication such as automobile telegram and in-company communication. It is. CELP (Code E) is used as a high-quality speech coding method at a bit rate of 8 kbps or less.
xcited Linear Prediction) system is known.

[0003] This CELP system is based on M & A, AT & T Bell Labs.
Code-Excited Linea by R, Schroeder and BSAtal
r Prediction (CELP) “High-Quality Speech at Very
LowBit Rates "Proc.ICASSP; 1985, pp. 937-939 (Reference 1), which has attracted attention as a method of synthesizing high-quality speech, such as improving quality and reducing the amount of calculation.
Various considerations have been made. The feature of the CELP system is L
The drive signal of the PC (Liner Predictive Coding) synthesis filter is stored in the codebook as a drive signal vector, and the optimum drive signal vector is evaluated from the codebook while evaluating the error between the synthesized speech signal and the input speech signal. The point to search.

FIG. 5 is a block diagram of a speech coding apparatus using the latest CELP method. In the figure, a sampled audio signal sequence as an input signal is input terminal 60
0 is input in frame units. When the frame is composed of L signal samples and the sampling frequency is 8 kHz, L = 160 is generally used. Although not shown in FIG. 5, prior to the search for the drive signal vector, LPC analysis is performed on the input L-sample audio signal sequence, and LPC prediction parameters ｛α _i , i = 1, 2,. p｝
Is extracted. This LPC prediction parameter α _i is LP
It is supplied to the C synthesis filter 630. Note that p depends on the prediction, and p = 10 is generally used. The transfer function H (z) of the LPC synthesis filter 630 is given by [Equation 1].

[0005]

(Equation 1)

Next, a process of searching for an optimal drive signal vector while synthesizing an audio signal will be described. First,
From the audio signal of one frame input to the input terminal 600, the subtracter 610 generates a synthesis filter 630 for the previous frame.
The effect of the internal state on the current frame is subtracted. The signal sequence obtained from the subtractor 610 is divided into four subframes, and becomes a target signal vector of each subframe.

[0007] A drive signal vector which is an input signal of the LPC synthesis filter 630 is obtained by multiplying a drive signal vector selected from the adaptive codebook 640 by a predetermined gain in a multiplier 650 and a white noise codebook 710. The noise vector obtained by multiplying the noise vector multiplied by a predetermined gain by the multiplier 720 is added by the adder 660.

Here, the adaptive codebook 640 is described in Reference 1.
Is performed by closed loop operation or analysis by synthesis (Analysis by Synthesis). For details, refer to WBKleijin DJ Krasinski and
RHKetchum, "Imptoved Speech Quality and Efficient
Vector Quantization in CELP ", Proc. ICASSP, 1988, pp.
155-158 (Reference 2). According to Document 2, the drive signal of the LPC synthesis filter 630 is transmitted over the pitch search range a to b (a and b are sample numbers of the drive signal, and usually a = 20, b = 147).
By delaying one sample at a time at 70, a drive signal vector for the pitch period of a to b samples is created and stored as a codeword in the adaptive codebook.

When searching for an optimal drive signal vector, the code words of the drive signal vector corresponding to each pitch period are read one by one from the adaptive codebook 640 and multiplied by a predetermined gain by a multiplier 650. . Then, a noise vector selected from the white noise codebook 10 and a product obtained by multiplying the noise vector by a predetermined gain by the multiplier 720 are added by the adder 660, and this signal is subjected to filter operation by the LPC synthesis filter 630, and synthesized. An audio signal vector is generated. The generated synthesized speech signal vector is subtracted by the subtractor 620 from the target signal vector. The output of the subtracter 620 is input to an error calculation circuit 690 via an audibility weighting filter 680, and a mean square error is obtained. The information of the mean square error is further supplied to the minimum distortion search circuit 7
00 and its minimum value is detected.

The above process is performed on the codewords of all the drive signal vectors in the adaptive codebook 640, and the minimum distortion search circuit 700 obtains the codeword number that gives the minimum value of the mean square error. The gain multiplied by the multiplier 650 is also determined so that the mean square error is minimized.

Next, an optimum white noise vector is searched for in the same manner. That is, the code words of the noise vector are read out one by one from the white noise codebook 710, multiplied by the gain in the multiplier 720, and the noise vector selected from the white noise codebook 10 is multiplied by the multiplier 7.
20 and the result of multiplication by a predetermined gain is added by an adder 660, and the result is input to an error calculation circuit 690 through a filter operation in an LPC synthesis filter 630, further through an audibility weighting filter 680, and A signal vector is generated, and an error calculation circuit 690 calculates a mean square error with respect to the target vector for all noise vectors. Then, the minimum distortion search circuit 700 obtains the number and gain of the noise vector that gives the minimum value of the mean square error. Note that the auditory weighting filter 680 is used to shape the spectrum of the error signal output from the subtractor 620 to reduce distortion perceived by humans.

As described above, the CELP method seeks an optimal drive signal vector that minimizes an error between a synthesized voice signal and an input voice signal, and thus synthesizes high-quality voice even at a low bit rate of about 8 kbps. can do.

On the other hand, when a speech coding apparatus (codec) is applied to mobile communication such as a car phone, communication is performed using radio waves, and transmission errors may occur depending on the surrounding environment. In order to reduce transmission, a transmission path coding technique is also used. Also, when the technology of the voice encoding device is applied to ATM communication and packet communication, congestion occurs in voice processed in packet form depending on the processing state of the packet, which causes cell discarding and packet discarding. Therefore, it is necessary to take measures to prevent the quality of the voice itself from deteriorating. In applying a transmission line code, a method of adding an error correction code only to a parameter having a high error sensitivity among coding parameters or a method of adding an error correction code having a different correction capability according to the magnitude of the error sensitivity is used. It has been conventionally used from the viewpoint of efficient allocation of channel codes.

In ATM communication or packet communication, a priority is given to a cell or packet to be transmitted (hereinafter referred to as priority), and a parameter having high error sensitivity is set and transmitted as a cell or packet with high priority.

In the CELP system shown in FIG. 5, the coding parameters to be transmitted include an LPC prediction parameter (maintenance of the synthesis filter 630), an adaptive codebook 640 number (hereinafter referred to as an index), a gain, and a white noise codebook 10. (Hereinafter referred to as an index) and a gain. Among these parameters, low-order LPC
The prediction parameter and the index and gain of the adaptive codebook are considered to be weak against errors (high error sensitivity), and although an error correction code is added, the index and gain of the white noise codebook are generally not protected. .

However, when an error occurs in the index or gain of the white noise codebook, the quality of the decoded speech is continuously deteriorated due to the error. This is because the driving signal of the synthesis filter is represented as the sum of the contribution of the adaptive codebook (the result of multiplication of the code vector and the gain) and the contribution of the white noise codebook, and the driving signal obtained by this sum is sequentially stored in the adaptive codebook. , In the next frame following the current frame.

[0017]

As described above, a conventional speech coding apparatus to which the CELP system is applied is provided with an adaptive codebook and a noise codebook, and performs speech coding using these. Therefore, when an error occurs in a parameter of the noise codebook which has not been subjected to the error correction processing in the related art, there is a problem that the quality cannot be reduced for a long time because the error cannot be reduced.

SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and has as its object to provide a speech coding apparatus in which the quality of a decoded speech is hardly deteriorated even if an error occurs in a parameter of a noise codebook. is there.

[0019]

In order to achieve the above object, according to the present invention, a codebook storing a plurality of drive signals and a noise codebook storing a plurality of noise parameters are referred to while referring to an input audio signal. An audio codebook that encodes the audio signal based on the first and second noise signals, wherein the adaptive codebook stores a plurality of first drive signals, a first noise codebook stores a plurality of first noise parameters, A second noise codebook storing a plurality of second noise parameters having a lower priority than the noise parameter of
First drive signal generating means for generating a first drive signal based on information from the adaptive codebook and the first noise codebook, and information from the adaptive codebook and the first and second noise codebooks Based on the first drive signal, a second drive signal generation means for generating a second drive signal, a means for delaying the generated first drive signal and storing the delayed first drive signal in an adaptive codebook, Transmission processing means for performing transmission processing so that the information has a smaller transmission error than the information based on the second drive signal.

[0020]

The noise parameter stored in the first noise codebook has a higher priority than the noise parameter stored in the second noise codebook.
The second drive signal is generated based on these noise code bits and information stored in the adaptive code book.
However, the first drive signal stored in the adaptive codebook is generated from the adaptive code and the first high-priority noise codebook.

Therefore, the second noise codebook having a low priority and in which an error is likely to occur is not used for updating the adaptive codebook, so that the influence of the error does not remain long and the speech quality is reduced.

[0022]

An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of a speech encoding apparatus according to one embodiment of the present invention.

In FIG. 1, an input audio signal is input to an input terminal 1.
00 is input to the frame buffer 101. In the frame buffer 101, an input audio signal sequence is cut out in units of L samples and stored as a signal for each frame. L is usually about 160. The input audio signal sequence for each frame from the frame buffer 101 is LP
It is supplied to the C analysis circuit 102 and the weighting filter 106.

The LPC analysis circuit 102 converts the input speech signal into an LPC (Linear Predi
ctive Coding (linear predictive coding)
The PC prediction coefficient {α _i , i = 1, 2,..., P} or the reflection coefficient {k _i , i = 1, 2,. The extracted prediction coefficient or reflection coefficient is encoded by the encoding circuit 103 with a predetermined number of bits, and then used by the weighting synthesis filters 107, 112, 122, and 152.

The weighting filter 106 weights an input speech signal sequence when searching for a driving signal vector of a synthesis filter from the adaptive codebook 110 and the noise codebooks 120 and 150. The transfer function H (z) of a general synthesis filter is described by the above [Equation 1]. At this time, the transfer function W (z) of the weighting filter 106 is represented by [Equation 2].

[0026]

(Equation 2) Here, γ is a parameter for controlling the weighting strength (0 ≦ γ ≦ 1).

The weighting synthesis filters 107, 112, 1
Reference numerals 22 and 152 each include a cascade connection of a synthesis filter of a transfer function HW (z) and a weighting filter of a transfer function W (z), and the transfer function HW (z)
Is described by [Equation 3].

[0028]

(Equation 3)

As in this embodiment, the weighting filter 106
Is used, it is possible to reduce the coding distortion on the auditory perception. Further, in the present embodiment, the weighting filter 106 is provided outside the drive signal vector search loop, and as a result, the amount of calculation required for the search is greatly reduced.

Further, the weighting synthesis filters 112, 1
The weighting synthesis filter 1 having an initial memory so that the first and second signals do not affect the search for the drive signal vector.
07 is provided. This weighting synthesis filter 10
7 is the last weighted synthesis filter 11 of the previous frame.
2, 122, and 152 have the internal state held as the initial state.

Then, a zero-input response vector of the weighting synthesis filter 107 is created, and the zero-input response vector is subtracted from the output of the weighting filter 106 in the subtracter 108. Thereby, the weighting synthesis filter 11
The initial states of 2, 122 and 152 can be set to zero, and the search for the drive signal vector can be performed without considering the influence of the previous frame. The above processes are all performed in frame units. Next, M frames (usually M =
The process of 4) of searching for a drive signal vector divided into subframes and performed in subframe units will be described.

The search for the drive signal vector is performed in the order of the adaptive codebook 110 and the noise codebooks 120 and 150. First, the pitch period j from the adaptive codebook 110
The drive signal vector X _j corresponding to
L / M = K) are sequentially read, and a multiplier 111 multiplies X _j by a predetermined gain β.
To supply. The weighting synthesis filter 112 performs a filtering operation to generate a synthesized speech vector.

On the other hand, the input audio signal read from the frame buffer 101 is weighted by the weighting filter 106, and the influence of the previous frame is subtracted by the subtracter 108. Using the audio signal vector Y output from the subtracter 108 as a target vector, the subtractor 113 calculates an error vector E _j with the synthesized audio vector from the weighting synthesis filter 112. Then, the square error calculation circuit 114 calculates the sum of squares of the error _{ E _j }, the minimum value of the _{ E _j } and the index j given by the minimum value.
Is detected by the minimum distortion search circuit 115. The index j is given to the adaptive codebook 110.

More specifically, the error vector E _j is represented by, for example, [Equation 4]. This error vector {E _j } is expressed as β
By substituting into zero by partial differentiation, the minimum value of {E _j } when β is optimized is expressed by [Equation 5]. Where β
Is a gain given by the multiplier 111.

[0035]

(Equation 4)

[0036]

(Equation 5)

Where {X} is the square norm, (X, Y)
Represents an inner product, and H is an impulse response matrix of a weighting synthesis filter (transfer function: HW (z)) given by [Equation 6].

[0038]

(Equation 6)

As apparent from [Equation 5], the search for the drive signal vector from the adaptive codebook 110 calculates the second term on the right side of [Equation 5] for all codewords _Xj , and the maximum is This is performed by detecting an index j that becomes

When the optimal drive signal vector X _opt is searched from the adaptive codebook 110 in this way, the output of the weighting synthesis filter 112 corresponding to X _opt is subtracted from the target vector Y by the subtractor 113, and this subtraction is performed. The output of the unit 113 is used as a target vector for noise vector search from the noise codebook 120. Noise codebook 1
The search for the noise vector from C.20 is also performed by the adaptive codebook 1
10 can be performed in exactly the same way as the search for the drive signal vector. Further, a search for a noise vector from the noise codebook 150 can be similarly performed. The code vector obtained by the search from the noise vector 120 is N _opt , the code vector obtained by the search from the noise code vector 150 is M _opt , and the noise code vector 150
And the code vector M _opt obtained by the search from
The drive signal vector X of the synthesis filter is expressed as X = β · X _opt + g · N _opt + f · M _opt . Here, β, g, and f are multipliers 11 respectively.
1, 121 and 151 are gains given to the drive signal vector and the noise vector searched from the adaptive codebook 110 and the noise codebooks 120 and 150. Next, the drive signal vector is applied to the adaptive codebook 1
The method of storing the data in the storage unit 10 will be described.

The drive vector X _s stored in the adaptive codebook 110 is expressed by the following formula:
And a code vector obtained by searching the noise codebook 110. X _s = β · X _opt + g · N _opt

The above addition is performed by the adder 130, and the result is supplied to the delay circuit 131. In the delay circuit, the drive signal vector is set to a pitch search range a to b (a, b
Is the sample number of the drive code, usually a = 20, b =
By delaying one sample at a time over 147), a drive signal vector for the pitch period of a to b samples is created, and this is stored in the adaptive codebook 110 as a codeword.

The bit string of the coded parameter obtained by the above processing is converted by the bit string
Class 1 bit string with high error sensitivity and high priority and Class 2 with low error sensitivity and low priority
Is divided into bit strings. FIG. 2 shows an example of coding parameter classification.

Next, error correction coding is performed by the error correction encoder 170 for the class 1 bit string, and error correction coding is performed by the error correction encoder 171 for the class 2 bit string. At this time, the error correction code for class 1 uses a code having higher correction ability and detection ability than the error correction code for class 2. For example, Reed-Solomon code for class 1,
A BCH code is used for class 2. There is also a method of not adding an error correction code to class 2.

It should be noted that in the present invention, as can be seen from FIG.
Code and noise codebook 120 output from
Are protected by an error correction code having a high correction capability. Then, as can be seen from the configuration of FIG. 1, using only the index and gain of the adaptive codebook 110 with high priority and the index and gain of the noise codebook 120 protected by the error correction code with high correction capability, Since the drive codes that are sequentially stored in the book are created, even if there is a transmission error, it is possible to achieve encoding with less deterioration in voice quality.

The outputs of error correction encoders 170 and 171 are distributed by interleaver 180 over a plurality of frames. This prevents the coding parameters from being erroneously bursty.

Next, FIG. 3 shows an embodiment of a speech decoding apparatus for decoding encoded speech in the speech encoding apparatus shown in FIG. The outline of this apparatus will be described with reference to FIG.

The bit string of the parameter encoded in FIG. 1 is input from input terminal 200 to deinterleaver 201. The deinterleaver 201 performs a formal rearrangement of the bit string to form a frame unit.
As a result, the class 1 bit string is
, The class 2 bit string is input to the error correction decoder 211. The input bit string is subjected to error correction decoding processing by the respective error correction decoders 210 and 211, and is configured as a coding parameter by the bit string integrator 220. The encoding parameters configured by the bit string integrator 220 are as follows:
The signals are input to the decoders 230, 231, 232, and 233, and are respectively decoded.

On the other hand, the bit string from the bit string integrator 220 is stored in the adaptive codebook 260 and the noise codebook 24.
0, 250 are input. And noise code book 2
The outputs of 40 and 250 are input to multipliers 241 and 251 and multiplied by the decoded outputs of decoders 233 and 232. The multiplied value is input to the heaters 262 and 263, and is used for processing for generating a drive signal. Based on the bit string input to adaptive codebook 260, the driving signal stored therein is input to multiplier 261;
The signal is multiplied by the decoded signal output from the decoder 231. The output of the adder 262 is input to the delay unit 263, and after being delayed by one sample over the pitch search range a to b, is input to the adaptive codebook 260.
Then, a loop-like process is performed. The output of the adder 263 is input to the synthesis filter and filtered based on the output signal of the decoder 230, and the output is input to the post filter 271. In the post filter 271, the spectrum is shaped by the signal of the decoder 230, and after the audibility is suppressed, the output is output from the output terminal 272.

In the block described above, a drive signal is created based on the decoded index and gain of the adaptive codebook 260 and the index and gain of the noise codebooks 240 and 250. The code vector, index, and gain of the adaptive codebook are X, I, and β, respectively.
The code vector, index, and gain of the noise codebook 250 are N and j, and the code vector, index, and gain of the noise codebook 240 are M, k,
Assuming that f, the drive signal X is expressed by the following equation. X = β · X _i + g · N _j + f · M _k FIG. 4 is a block diagram of a speech encoding apparatus according to another embodiment in which the present invention is applied to ATM communication.

The differences from the previous embodiment shown in FIG.
After classifying the bit string into class 1 and class 2 in bit string classifier 160, cell forming section 190 converts the bit string of class 1 into a priority cell and the bit string of class 2 into a non-priority cell.
Is to make a cell.

In the embodiment described above, the codeword of the noise codebook is created from a white noise sequence.
The codeword may be given as a sum of a plurality of basis vectors. An encoding method having such a codebook is VS
This method is called the EP (Vector Sum Excited Linear Prediction) method, and is called “VE” by Ira A.Gerson and Mark A.Jasiuk.
CTOR SUM EXCITED LINEAR PRIDICTION (VSELP) SPEECH
CODING AT 8KBPS ”, International Conterence on Ac
oustiecs, Speech and Signal Processing, April 1990. The use of this method has the effects of significantly reducing the amount of calculation required for searching for a noise codebook and increasing the resistance to transmission errors. Also, as used in this document, the noise codebooks 120, 15
In the search for 0, each codeword is orthogonalized with the target vector, and the gain is vector-quantized, so that the quality of synthesized speech can be improved.

[0053]

As described above, according to the present invention, the generation of the first drive signal which is repeatedly used is the highest priority among the information stored in the adaptive codebook and the noise codebook. Since it is performed from the information stored in the first noise codebook, even if the information stored in the second low-priority noise codebook is erroneous, the effect is not affected by the voice quality, It is possible to provide a speech encoding device with less degradation in speech quality.

[Brief description of the drawings]

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

FIG. 2 is a diagram showing one class of parameters used for speech encoding according to the present invention.

FIG. 3 is a block diagram of a speech decoding apparatus according to one embodiment of the present invention.

FIG. 4 is a diagram showing another embodiment of the present invention.

FIG. 5 is a diagram showing a conventional example.

[Explanation of symbols]

 Reference Signs List 100 audio signal input terminal 101 frame buffer 102 LPC analysis circuit 103 encoding circuit 104 pitch analysis circuit 106 weighting filter 107 weighting synthesis filter 110 adaptive codebook 112 weighting synthesis filter 114 square error calculation Circuit 115: Minimum distortion search circuit 120: Noise codebook 122: Weighting synthesis filter 124: Square error calculation circuit 125: Minimum distortion search circuit 131: Flip sampling circuit 132: Delay circuit 133: Thinning circuit 140: Gain encoding circuit 141 ... Gain encoding circuit 152 ... Weighting synthesis filter 154 ... Square error calculation circuit 155 ... Minimum distortion search circuit 160 ... Bit string classifier 170 ... Error correction encoder 171 ... Error correction encoder 180 ... Interleaver 181 ... Output Child

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁷ , DB name) G10L 19/12

Claims (1)

(57) [Claims] 1. A speech encoding apparatus for encoding a speech signal based on a codebook storing a plurality of drive signals and a noise codebook storing a plurality of noise parameters while referring to the input speech signal, An adaptive codebook storing a plurality of drive signals of the first type, a first noise codebook storing a plurality of the first noise parameters, and a plurality of second noise parameters having a lower priority than the first noise parameter are stored. A second noise codebook; first drive signal generating means for generating the first drive signal based on information from the adaptive codebook and the first noise codebook; A second drive signal generating means for generating a second drive signal based on information from the first and second noise codebooks; Means for delaying the first drive signal and storing it in the adaptive codebook; and transmitting processing so that information based on the first drive signal has a smaller transmission error than information based on the second drive signal. And a transmission processing unit.