JPH0854900A - Coding/encoding system by vector quantization - Google Patents

Abstract

PURPOSE:To perform high quality vector quantization which produces a small amount of errors while maintaining the amount of information in a code book and between a coding device and an encoding device small. CONSTITUTION:The coding/encoding system performs vector quantization employing a code book which stores a correspondence list of signal patterns within a prescribed segment and indexes. The system is provided with a main code book 7 and an error code book 10. A subtractor 8 computes an error vector between a main code vector and a quantized vector which is vector quantized in a main vector quantizing section 6 by referring to the book 7. The error vector is again vector quantized in an error vector quantizing section 9 by referring to the book 10. If the error vector is detected to be larger than a prescribed value by an error amount detection section 11, a main index IDX as well as an error index EIDX are transmitted. A main code book updating section 13 updates the contents of the book 7 employing a new vector which is produced by modifying the main code vector by the error code vector.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an encoding / decoding system by vector quantization for highly efficiently compressing voices, images, etc. using a coatbook for transmission or recording.

[0002]

2. Description of the Related Art Conventionally, vector quantization has been known for compressing voices and images with high efficiency for transmission or recording. In vector quantization, the signal waveform to be quantized is cut out for one regular section, and the waveform pattern in this section is expressed by one code (index). Therefore, various waveform patterns are stored in advance and an index is added to each waveform pattern. Various waveform patterns are vector information expressed by the number of dimensions of the number of samples. These vector information are called code vectors, and the correspondence table between these code vectors and indexes is called a codebook.

[0003] For example, CELP ("Code-Excited Linear Prediction:") which is known as one of audio coding systems.
High Quality Speech at Very low Bit Rate "MRSch
Roeder, BSAtal ICASSP, 1985) divides a prediction residual component of speech into data of about 40 samples called subframes, and applies vector quantization to each subframe. However, the codebook used for this vector quantization is not optimized for any particular individual as it will be average of many speakers. For this reason, the error between the quantized vector and the codebook vector is large, and good quality speech cannot be transmitted. If the number of code vectors stored as a codebook is large, the sound quality is improved, but on the other hand, the amount of codebook is large and the cost is increased, and the information amount of the index is also increased, so that the transmission amount is also increased. There is also a problem that it takes time to search the codebook.

Therefore, Japanese Patent Laid-Open No. 186836/1990 discloses a vector quantizer which minimizes the occurrence of quantization distortion with a small amount of codebook information. This vector quantizer also transmits the quantized vector to the decoding device side when the minimum distortion at the time of vector quantization is larger than a predetermined value, and selects the quantized most recently selected in the codebook. By rewriting as a representative vector (code vector), the contents of the codebook are adaptively updated according to the characteristics of the quantized vector.

[0005]

However, in the above-mentioned conventional vector quantizer, when the rewriting of the codebook occurs, the new quantized representative vector itself has to be transmitted or recorded. There is a problem that the amount increases. This is a major drawback when applied to speech coding / decoding for mobile phone devices using vector quantization with limited transmission information capacity.

The present invention has been made in view of the above problems, and it is possible to suppress the amount of information in the codebook and between the encoding device and the decoding device to a small amount, and to perform high-quality vector quantization with few errors. It is an object of the present invention to provide an encoding / decoding method by vector quantization capable of realizing the following.

[0007]

A coding / decoding method by vector quantization according to the present invention specifies main code vectors indicating some typical signal patterns in a predetermined section and the main code vectors respectively. A main codebook storing a correspondence table with a main index, and some typical patterns of error vectors showing the difference between the quantized vector to be quantized and the main code vector selected from the main codebook. And an error codebook that stores a correspondence table between error code vectors that indicate the error code vectors and error indexes that specify the error code vectors respectively, are provided in both the encoding device and the decoding device, and in the encoding device, , Select the main code vector in the main codebook that best matches the quantized vector, While outputting the index, the difference between the quantized pattern and the selected main code vector is calculated as an error vector, and when the error vector is larger than a predetermined value, the error that best matches the error vector. The error code vector in the codebook is selected, the error index is output, and the decoding device obtains the main code vector corresponding to the main index obtained on the encoding device side by referring to the main codebook. Along with, when the error vector is larger than a predetermined value, the error code vector corresponding to the error index obtained on the encoding device side is obtained by referring to the error code book, and the obtained main code vector is Characterized by decoding the quantized vector by synthesizing with an error code vector That.

The encoding device and the decoding device may further include main codebook rewriting means for appropriately correcting the main code vector in the main codebook by the selected error code vector. . Also,
The main codebook and the error codebook may be shared as one codebook, and one code vector may be used as both the main code vector and the error code vector.

[0009]

According to the present invention, when the quantized vector is vector quantized by the main codebook, the amount of error between the quantized vector and the main code vector is detected. Further vector-quantizes the error vector by an error codebook.
Therefore, the quantization distortion component in the vector quantization is also information-compressed by the vector quantization, and the error amount is increased without increasing the information amount of the codebook and the information amount passed between the encoding device and the decoding device. It is possible to realize high-quality vector quantization with less number of bits.

When the amount of quantization distortion during vector quantization is large, the contents of the codebook are updated with the error-corrected vector so that the contents of the codebook are adaptive according to the characteristics of the quantized vector. Can be changed to Then, even in this case, according to the present invention, since only the error index, that is, the information of the vector-quantized distortion component, is passed from the encoding device side to the decoding device side, the new code vector itself is transmitted or Compared with the conventional method of recording, the amount of information transmitted or recorded can be significantly reduced.

[0011]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 shows a CELP according to an embodiment of the present invention.
FIG. 2 is a block diagram showing a speech signal coding apparatus including vector quantization according to the method, and FIG. 2 also showing a decoding apparatus.
In FIG. 1, the audio signal input to the encoding device is A
After being A / D converted by the / D converter 1, it is supplied to the long-term predictor 2. The long-term predictor 2 divides the input digital speech signal into frames of, for example, 160 samples, analyzes each frame, and encodes / multiplexes long-term predictive coefficients such as percoll coefficients, gains, and pitches. Output to 3. Further, the long-term predictor 2 subtracts the signal linearly predicted by the long-term prediction coefficient from the input digital audio signal to obtain a long-term prediction residual component, and outputs this to the frame division unit 4. The frame division unit 4 uses the frame timer 5 to divide the long-term prediction residual component of 160 samples into subframes of, for example, 40 samples each, and the main vector quantization unit 6 outputs the 40-dimensional prediction residual vector to the main vector quantization unit 6. Output. In CELP, this prediction residual vector becomes a quantized vector to be vector-quantized. The main vector quantization unit 6 normalizes the prediction residual vector so that the magnitude of the vector is 1, and determines the main code vector closest to the normalized prediction residual vector in distance to the main codebook 7 Search from and its main index I
The DX and the normalization coefficient are output to the encoding / multiplexing unit 3.

The main codebook 7 has a structure as shown in FIG. 3, for example. That is, in the main codebook 7,
For example, 51 selected from various languages, races and genders
Main code vectors of two representative prediction residual components are stored in advance. Further, the main codebook 7 stores a main index IDX that identifies each main code vector in a form corresponding to each main code vector. If the number of main code vectors is large, the quantization error will be small, and high-quality encoding / decoding can be realized, but on the other hand, the memory capacity must be increased and the cost will increase, and the search It also takes time. Also, for example, 51
If there are two types of main code vectors, the main index I
Although 9-bit information may be transmitted as DX, the number of bits to be transmitted increases as the types of main code vectors increase.

When the pattern vector forming the main codebook 7 is expressed as a two-dimensional vector on a two-dimensional plane for simplicity, it becomes like a black circle in FIG. 4 (a). Here, the numerical value written in black circles is the main index of each main code vector. Now, when a vector Vi indicated by x in the figure as a prediction residual vector to be quantized is input to the vector quantizing unit 6, since this vector Vi is the closest to the vector V12, the main index 12 is the main vector quantizing unit 6. Will be selected as the quantization result in.

At this time, the error between the prediction residual vector and the selected main code vector (error between X and V12) is calculated by the subtractor 8. The output of the subtractor 8 is given to the error vector quantizer 9 as an error vector. The error vector quantization unit 9 normalizes the error vector so that the magnitude of the vector becomes 1, and searches the error code book 10 for an error code vector that is closest in distance to the normalized error vector. Its error index E
The IDX and the normalization coefficient are output to the encoding / multiplexing unit 3. The error codebook 10 has the same structure as the main codebook 7. A large codebook is effective for improving the correction efficiency, but since the main codebook 7 is adaptively rewritten in this system, it is not necessary to prepare so many different error code vectors, and 40-dimensional (sample 40), it is sufficient to prepare 40 error code vectors. In this case, the number of patterns that can be represented by the combination of the main codebook 7 and the error codebook 10 is 512 × 40 = 2048.
Since there are 0 patterns, it is equivalent to storing 20480 patterns in the conventional codebook with 552 patterns.

The error vector from the subtractor 8 is also given to the error amount detector 11. When the magnitude of the error vector is larger than a predetermined value, the error amount detection unit 11 sets the error transmission flag for transmitting or recording the error component to “1”.
To When the error transmission flag is “0”, that is, when the error component is less than the predetermined value, the encoding / multiplexing unit 3
The long-term prediction coefficient, IDX, its normalization coefficient, and the error transmission flag are encoded and multiplexed and transmitted to the transmission line or recorded on a recording medium. When the error transmission flag is "1", EIDX and its normalization are added to these information. A coding coefficient is added and encoded and multiplexed to be transmitted to a transmission line or recorded in a recording medium.

On the other hand, on the decoding device side, as shown in FIG. 2, the received or read signal is supplied to the de-encoding / de-multiplexing unit 21 and de-encoded / de-multiplexed to Long-term prediction coefficient and normalization coefficient, and the main index I
It is divided into DX, error index EIDX, and error transmission flag. The long-term prediction coefficient and the normalization coefficient are supplied to the long-term prediction synthesis unit 22. Further, the decoding device side is also provided with the main codebook 24 and the error codebook 26 having the same contents as the encoding device, and the main vector dequantization unit 23 uses the supplied main index IDX to the main codebook 24. , The error vector dequantization unit 25 selects the error code vector by referring to the error code book 26 from the supplied error index EIDX.
As a result, the main codebook 24 and the error codebook 2
The main code vector and the error code vector respectively output from 6 are added by the adder 27 and supplied to the long-term prediction synthesis unit 22. The long-term prediction synthesis unit 22 uses the adder 2
The added value of the main code vector and the error code vector supplied from 7 is returned to the original signal level by the normalization coefficient, and this signal and the signal obtained by linear prediction from the long-term prediction coefficient are combined. This combined signal is the D / A converter 28.
Is D / A converted by and output as an audio signal.

Next, the updating process of the main codebooks 7 and 24 will be described. In the encoding device, the main code vector selected by the main codebook 7 and the error codebook selected by the error codebook 10 are added by an adder 12
Is added by The output of the adder 12 has a value closer to the prediction residual vector input to the main vector quantization unit 6 than the main code vector selected by the main vector quantization unit 6. On the other hand, in the decoding device, the adder 2
The output of 7 is the same as the output of the adder 12 of the encoder. Then, the outputs of these adders 12 and 27 are replaced with the selected main code vector in the main codebooks 7 and 24 as a new main code vector when the error transmission flag is "1". FIG. 4B shows an example in which the main code vector V12 in FIG. 4A and the main code vector newly generated from the error code vector are replaced. Further, a new main code vector may be replaced with a main code vector having the lowest selection frequency per predetermined time.

The contents of the codebook can be adaptively changed according to the characteristics of the speaker by updating the main codebook 7 as described above. When the characteristics change suddenly, the content of the learned codebook may be rather inconvenient. In this case, the contents of the codebooks 7 and 24 may be reset to the initial contents by an initialization signal or the like. The initial data of the codebook is other R
Although it may be stored in the OM or the like, only the rewritten vector information is stored in the empty areas of the main codebooks 7 and 24, and the contents are restored based on the initialization signal. Is also good.

5 and 6 are diagrams for explaining an encoding / decoding system according to another embodiment of the present invention. FIG. 5 is a block diagram showing an encoding device and FIG. 6 is a block diagram showing a decoding device. Is. The system of this embodiment is different from the system of the above-described embodiment in that the main codebook and the error codebook are shared as one codebook 14 and 30, and the code by the codebook rewriting units 15 and 31. The point is that rewriting weighting units 16 and 32 that give weighting regarding whether or not rewriting is possible are added when the book is updated.

In this embodiment, the main code vector and the error code vector are shared, and one code vector can be used as both the main code vector and the error code vector. If so, it becomes 5 with the conventional code vector.
This is equivalent to storing 12 × 512 = 262144 code vectors. It is desirable that the codebooks 14 and 30 include primitive vectors, such as vectors in the X, Y, and Z directions, if they are three-dimensional vectors. This is because such a vector is necessary as an error code vector for correcting the error. In other words, the code vector usable as such an error code vector is the code book rewriting unit 15, 31.
It is better not to rewrite. Therefore, the rewriting weighting unit 1
6 and 32 weight each code vector so as to suppress the rewriting of such a primitive code vector. In addition, if the primitive code vector is written to the beginning of the code book 14 as much as possible,
The search time for the error code vector can be shortened.

In the above embodiments, the vector quantization of the prediction residual vector of the voice signal is taken as an example, but the same applies to the vector quantization of the voice signal itself or the image signal. Needless to say.

[0022]

As described above, according to the present invention,
When the quantized vector is vector quantized by the main codebook, the amount of error between the quantized vector and the main code vector is detected. Since the vector is quantized by the book, the information about the quantized distortion component in the vector quantization is further compressed by the vector quantization, and is transferred between the codebook information amount and the coding device and the decoding device. It is possible to realize high-quality vector quantization with a small error without increasing the amount of information.

Further, even when the contents of the codebook are updated to the error-corrected vector when the amount of quantization distortion during vector quantization is large, only the error index, that is, the information on the vector-quantized distortion component, is encoded. Since it suffices to pass the information from the encoding device side to the decoding device side, the amount of information transmitted or recorded can be greatly reduced as compared with the conventional method of transmitting or recording the new code vector itself.

[Brief description of drawings]

FIG. 1 is a block diagram of an encoding device by vector quantization according to an embodiment of the present invention.

FIG. 2 is a block diagram of a decoding device using the same vector quantization.

FIG. 3 is a diagram for explaining the contents of a codebook in the encoding device and the decoding device.

FIG. 4 is a diagram schematically showing the contents of code vectors before and after updating the codebook.

FIG. 5 is a block diagram of an encoding device by vector quantization according to another embodiment of the present invention.

FIG. 6 is a block diagram of a decoding device using the same vector quantization.

[Explanation of symbols]

1 ... A / D converter, 2 ... Long-term prediction unit, 3 ... Encoding / multiplexing unit, 4 ... Frame division unit, 5 ... Frame timer, 6 ...
Main vector quantizer, 7, 24 ... Main codebook, 8 ...
Subtractor, 9 ... Error vector quantizer, 10, 26 ... Error codebook, 11 ... Error amount detector, 12, 27 ... Adder, 13, 29 ... Main codebook rewrite unit, 14, 30 ...
Code book, 15, 31 ... Code book rewriting section, 1
6, 32 ... Rewriting weighting section, 21 ... Decoding / demultiplexing section, 22 ... Long-term prediction combining section, 23 ... Main vector dequantizing section, 25 ... Error vector dequantizing section, 28 ... D / A conversion vessel.

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Internal reference number FI Technical indication location G10L 9/14 GH03M 7/30 B 9382-5K H04N 7/32

Claims (3)

[Claims] 1. A main codebook, which stores a correspondence table of main code vectors showing some typical signal patterns in a predetermined section and main indexes specifying the respective main code vectors, and An error code vector showing some typical patterns of error vectors showing the difference between the power-quantized vector and the main code vector selected from the main codebook, and an error index specifying each of these error code vectors. And an error codebook storing a correspondence table of, both the encoding device and the decoding device are provided, and in the encoding device, the main code in the main codebook that most matches the quantized vector. The vector is selected and its main index is output, and the quantized vector and the selected The difference from the main code vector is calculated as an error vector, and when this error vector is larger than a predetermined value, the error code vector in the error code book that most matches the error vector is selected and its error index is output. In the decoding device, the main code vector corresponding to the main index obtained on the encoding device side is obtained with reference to the main codebook, and when the error vector is larger than a predetermined value, the code The error code vector corresponding to the error index obtained on the digitizer side is obtained by referring to the error code book, and the obtained main code vector and error code vector are combined to decode the quantized vector. A coding / decoding method by vector quantization characterized by the following. 2. A codebook, which stores a correspondence table of main code vectors showing some typical signal patterns in a predetermined section and main indexes specifying the respective main code vectors, and to be quantized. Main vector quantizing means for comparing the quantized vector and each main code vector of the main codebook and selecting the main code vector that best matches them, and outputting the main index, the quantized vector and the An arithmetic means for calculating a difference from the main code vector selected by the main vector quantization means as an error vector, an error code vector showing some typical patterns of this error vector, and those error code vectors are respectively specified. An error codebook that stores a correspondence table with error indexes, and Error amount detecting means for detecting that the issued error vector is larger than a predetermined value; and the error vector and the error code when the error vector detecting means detects that the error vector is larger than a predetermined value. An encoding device by vector quantization, comprising: an error vector quantization means for comparing each error code vector of a book, selecting an error code vector that best matches them, and outputting the error index thereof. 3. A codebook, which stores a correspondence table of main code vectors showing some typical signal patterns in a predetermined section and main indexes specifying the respective main code vectors, and to be quantized. An error code vector showing some typical patterns of error vectors showing the difference between the quantized vector and the main code vector selected from the main codebook, and an error index specifying each of these error code vectors An error codebook storing a correspondence table, a main vector dequantization means for obtaining a main code vector corresponding to a given main index by referring to the main codebook, the quantized vector and the main code vector Error code vector corresponding to the error index given when the error amount between and is large Error vector dequantization means for obtaining the error codebook with reference to the error codebook, and a main code vector obtained by the main vector dequantization means and an error code vector obtained by the error vector dequantization means And a calculation means for outputting the result of the inverse quantization of the quantized vector.