JPH09261065A - Quantization device, inverse quantization device and quantization and inverse quantization system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To execute the vector quantization of a high quality without increasing an information transmitting quantity by providing a coding notebook with plural sub-coding notebooks with different vector coded words, a transmission sub-coding notebook selection means, first and second vector quantization means and a comparing means. SOLUTION: An inputted vector 5 is inputted to a vector quantization means 12 and a predictive error vector quantization means 13. The means 12 calculates a Euclidean distance the vector coded word within a coded word set 15 selected by a sub-coding notebook selection means 9 from the coding notebook 14 and the inputted vector 5 and outputs the code number 17 of an optimum vector coded word in it and a minimum Euclidean distance to a comparing means 21 as quantization distortion 18. Then at the time of the vector quantization of an input signal, two kinds of quantization of quantizing a present value and quantizing a value obtained by predicting from hourly preceding value are executed, and both of these quantigations are compared with each other and in addition, precise quantization can be then executed.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a quantizer and an inverse quantizer for quantizing and dequantizing an input vector obtained from an input signal using a codebook.

[0002]

[Prior art]

Conventional example 1. In FIGS. 14 and 15, an input vector obtained from an input signal such as speech is quantized by selecting a vector codeword with less distortion using a codebook containing many vector codewords for quantization. The conventional quantizer,
As a conventional dequantization device that performs dequantization, Japanese Patent Laid-Open No. 2-
This is disclosed in Japanese Patent No. 91700.

The operation of the quantizing device and the dequantizing device for quantizing and dequantizing the input vector obtained from the conventional input signal using a codebook will be described below. FIG.
FIG. 15 is a block diagram showing the overall configuration of a conventional quantization device, and FIG. 15 is a block configuration diagram showing the overall configuration of a conventional inverse quantization device.

In the figure, 101 is a quantizer, and 105
Is an inverse quantizer, 3 is an input signal, 4 is an input signal analyzing means, 5 is an input vector, 102 and 106 are codebooks, 1
Reference numerals 03 and 107 denote a codeword set A composed of vector codewords of M positive integers, 104 and 108 denote a codeword set B composed of NM prediction error vector codewords, 12 a vector quantization means, and 13 a prediction Error vector quantizer,
Reference numerals 17 and 19 are code numbers, 18 and 20 are quantization distortions, 2
1 is a comparison means, 22 is a past code number, 24 is a transmission path signal which is a selected code number, 30 is a control means, 31 is an inverse vector quantization means, 32 is an inverse prediction error vector quantization means, and 36 is Control signal, 37 is past decoding vector,
38 is a decoding vector, and 50 is a changeover switch.

First, in the quantizer 101, a digital audio signal sampled at, for example, 8 kHz is input signal 3
Is entered as In the input signal analysis means 4, the input signal 3 is divided into, for example, frame lengths of 20 ms, linear prediction analysis is performed, and an nth-order linear prediction parameter, which is spectral envelope information of voice, is obtained. (LSP), and the LSP is input to the vector quantization means 12 and the prediction error vector quantization means 13 as an n-dimensional input vector 5. The vector quantizer 12 calculates, for example, a Euclidean distance between the M vector codewords in the codeword set A103 and the input vector X, and the code number 17 of the optimum vector codeword giving the minimum distance among them. And the minimum Euclidean distance at that time is output to the comparison means 21 as the quantization distortion 18.

Further, the prediction error vector quantization means 1
3 is the past code number 22 transmitted up to the current processing time point.
The prediction value X _p of the current input vector 5 is obtained by using the quantization result of the past input vector represented by and the prediction error vector X _e between this prediction value X _p and the input vector X is calculated. . Then, similarly to the distance calculation of the vector code word, N−M prediction error vector code words and the prediction error vector X _e in the code word set B104.
Of the optimal prediction error vector codeword that calculates the Euclidean distance between
And the minimum Euclidean distance at that time is output to the comparison means 21 as the quantization distortion 20.

The comparing means 21 compares the two quantization distortions 18 and 20 inputted from the vector quantizing means 12 and the prediction error vector quantizing means 13 and predicts the code number giving a small quantizing distortion as a prediction error. In addition to outputting the past code number 22 to the vector quantizing means 13, the quantizing distortion having the smaller quantizing distortion is selected from the two quantizing distortions 18 and 20 and is also output as the transmission path signal 24. . The above is the outline of the quantization device in the conventional example.

Next, the operation of the conventional inverse quantizer will be described. In the inverse quantizer 105, the control means 30
Indicates that the code number 27 input as the transmission line signal 24 is
When designating a codeword in the codeword set A107,
When the code number 27 is output to the inverse vector quantization means 31 and the code word in the code word set B108 is designated, the code number 27 is output to the inverse prediction error vector quantization means 32 and the control signal is output. At 36, output control is performed by the changeover switch 50.

The inverse vector quantization means 31 comprises a control means 3
When the code number 27 is input from 0, the code number 27
The vector codeword in the codeword set A107 designated by is read and the decoded vector is output. When the code number 27 is input from the control means 30, the inverse prediction error vector quantization means 32 reads out the prediction error vector code word in the code word set B108 designated by the code number 27 and also decodes the past decoding vector 37. Is used to find the predicted value of the current input vector. Then, the prediction value of the input vector is added to the prediction error vector code word to generate a decoded vector, which is output as the decoded vector 38 via the changeover switch 50.

Conventional example 2. Further, as another conventional method provided with a plurality of codebooks, FIG. 16 shows an encoding device for compressing, transmitting and accumulating voice signal information.
This is disclosed in Japanese Patent Laid-Open No. 100. Hereinafter, the operation of the above-described conventional coding apparatus having a plurality of codebooks will be described. 16 is a block configuration diagram showing an overall configuration of a conventional encoding device, and FIG. 17 is a block configuration diagram showing an overall configuration of a decoding device corresponding to the encoding device of FIG. In the figure, 111 is an AD converter, 112 and 11
3 and 133 are linear prediction analyzers, 114 and 115 are linear predictors, 116 and 130 are a plurality of residual signal codebooks,
117 and 131 are multipliers, 118 and 132 are adders, 119 is a subtractor, 120 is an auditory weighting filter,
121 is a power calculator, 122 is an optimum codeword selector, 1
23 is a multiplexer, 124 is a fast Fourier transformer (hereinafter referred to as FFT), 125 is an inverse fast Fourier transformer (hereinafter referred to as inverse FFT), 126 is a spectrum inverter, 127 is a classifier, 128 is a transmission line, Reference numeral 129 is a demultiplexer, and 134 is a DA converter.

The AD converter 111 converts an analog voice signal input from an input signal into digital data, and the linear prediction analyzer 1 as an input vector S having a constant sample length.
12 and the subtractor 119. Linear prediction analyzer 11
2 performs linear prediction analysis on the digital data converted by the AD converter 111. Here, it is assumed that linear prediction analysis includes long-term prediction of spectrum proximity prediction and pitch prediction. The analysis is performed on a plurality of input vectors, and the linear prediction vector α as a prediction signal obtained as a result of the analysis is output to the linear predictors 114 and 115 and the multiplexer 123.

The linear predictor 115 is a linear predictive analyzer 11.
The residual signal of the input signal is calculated by being driven by the linear prediction vector α output from 2. If the linear prediction analyzer 112 has sufficiently performed linear prediction, the residual signal calculated by the linear predictor 115 is a zero point included in the audio signal, that is, a valley of the spectrum due to the anti-resonance point. including. The spectrum outline of the zero point described above is estimated by the following steps. The residual signal calculated by the FFT 124 is frequency analyzed and converted into a power spectrum (step S1). Spectrum inversion is performed by the spectrum inverter 126 for each power spectrum signal. here,
If the inverted spectrum component is S _i hat and the power spectrum component is S _i , S _i × S _i hat = 1. However, i = 0, ..., N-1, and N = FFT number of points (step S2). An inverse FFT is applied to the spectrum that has been inverted by the inverse FFT 125 to restore the time waveform (step S3).

The zero-type representation of the power spectrum is converted to the polar representation by the operations of the above steps S1 to S3. Then, linear prediction analysis is performed by the linear prediction analyzer 113 to extract the linear prediction vector α ′, and the extracted linear prediction vector α ′ is classified into several types by the classifier 127 using the vector quantization method. The index j as the command signal having the thus-obtained spectrum shape is output (step S4). After passing through each of the above steps S1 to S4, based on the spectrum shape index j output from the classifier 127, a codebook to be used is selected from a plurality of codebooks 116 stored in advance by the optimum codeword selector 122. Select. The spectral shape index j output from the classifier 127 is also output to the multiplexer 123.

A plurality of (for example, 256) gain-normalized residual signal waveforms (codewords) are stored in each of the codebooks constituting the plurality of codebooks 116. All codewords selected by the optimum codeword selector 122 are
It is amplified by the gain multiplier 117. Gain multiplier 1
The code word amplified in 17 is output to a synthesis filter including a linear predictor 114 and an adder 118.
The synthesis filter inputs the amplified codeword and outputs a synthesis signal vector S ′.

The subtractor 119 calculates an error between the combined signal vector S ′ output from the combining filter and the input vector S output from the AD converter 111, that is, an error. The error calculated by the subtractor 119 is perceptually weighted by the perceptual weighting filter 120. The power calculator 121 calculates a squared error of the perceptually weighted error output from the perceptual weighting filter 120, and outputs the squared error to the optimum codeword selector 122. The optimum codeword selector 122 selects the codeword and the gain so that the square error of the error calculated by the power calculator 121 is minimized, and the multiplexer 123 sets the index value index that identifies the selected codeword. Output to. The multiplexer 123 has a linear prediction vector α, a residual gain g,
The index j for selecting the codebook and the index value index for specifying the codeword are multiplexed and sent to the transmission line 128. The above is the outline of the encoding device including a plurality of codebooks in the conventional example.

The operation of the decoding device corresponding to the conventional encoding device shown in FIG. 16 will be described. The demultiplexer 129 decomposes the signal received from the transmission path 128 into a linear prediction vector α, a residual gain g, an index j of the residual spectrum shape, and an index value index that specifies the residual signal, and the decomposed residual The shape index j of the spectral shape and the index value index specifying the residual signal are output to a plurality of residual signal codebooks, the residual gain g is output to the adder 132, and the linear prediction vector α is output to the linear predictor 133.

Based on the shape index j output from the demultiplexer 129, a codebook to be used is selected from a plurality of codebooks 130 having the same contents as the encoding device. The codeword based on the selected index value index of the codebook is output to the multiplier 131. The multiplier 131 amplifies the codeword input by the received residual gain g and outputs it to the synthesis filter. The synthesis filter configured by a linear predictor having the received linear prediction vector α as a coefficient inputs the amplified codeword and outputs a reproduction signal vector.
The DA converter 134 inputs the reproduction signal vector output from the synthesis filter and converts it into an analog audio signal.

[0018]

In the first conventional example, the above-described conventional quantizing device and dequantizing device have one codebook, and the codes in the codeword set are adapted to the characteristics of the input signal. There is a problem that the composition ratio of the number of words cannot be changed, and the quantization or the dequantization cannot be efficiently performed by specializing the important feature part of the input signal. Further, in the above-described second conventional example of the coding apparatus including a plurality of codebooks, the codebook used for vector quantization has only a single quantization method, and input signals from a plurality of quantization methods are input. Quantization cannot be performed by switching the quantization method that suits the characteristics, and the information for specifying the codebook, which codebook was selected, must be transmitted, which increases the transmission amount. .

The present invention has been made to solve the above problems, and prepares a plurality of sets of sub-codebooks according to the characteristics of the power contained in the input signal, and selects and sets the subcodebooks for quantization and dequantization. By selecting, the selection information is
It is an object of the present invention to obtain a highly efficient quantizer, dequantizer, and quantized dequantizer system that also use other essential transmission information and reduce the amount of transmission information.

[0020]

A quantizer according to the present invention is a codebook in which a plurality of vector codewords matching the characteristics of an input signal are accommodated in advance.
Of the first vector code word corresponding to the vector quantizing means and the second code word corresponding to the second vector quantizing means in which the accommodating ratio of the second vector code word is changed. , Transmitting sub-codebook selecting means for specifying which of the plurality of sub-codebooks is to be selected based on the extracted information regarding the power of the input signal, and the sub-codebook selected based on the designation of the selection. First vector quantizing means for selecting a first vector code word adapted to an input signal by a code book, second vector quantizing means for selecting a second vector code word, and first vector quantizing means. The first vector code word selected in step 2 is compared with the second vector code word selected by the second vector quantization means to select and select a vector code word that produces a small quantization distortion. Enter the result as above And a comparing means for transmitting the output corresponding portion that outputs information regarding power.

The dequantization apparatus according to the present invention is a codebook in which a plurality of vector codewords adapted to the characteristics of an output signal are accommodated in advance. Configuring the above codebook with a plurality of sub-codebooks in which the number of vector codewords and the accommodation ratio of the second vector codeword corresponding to the second inverse vector quantization means are changed.
Receiving sub-codebook selecting means for extracting information on the power of the received input signal and designating which of the sub-codebooks to select based on the extracted information, and a sub-codebook selected based on the designation of the selection The first inverse vector quantization means for selecting the first vector code word by the designation in the input signal, the second inverse vector quantization means for selecting the second vector code word, and the first inverse vector The first vector code word selected by the quantizing means is compared with the second vector code word selected by the second inverse vector quantizing means to select a decoding vector that produces a small quantization distortion. And a control means for controlling so.

Alternatively, the quantizing means may be replaced with a codebook composed of a plurality of sub-codebooks, and the first vector codeword and the second vector codeword corresponding to the first vector quantizing means adapted beforehand to the characteristics of the input signal. A codebook accommodating a plurality of second vector codewords corresponding to vector quantization means, and a part of the codebook is cut out by changing the ratio of the number of words of the first vector codeword to the number of words of the second vector codeword. And a window moving means serving as a sub-codebook, and the transmission sub-codebook selecting means extracts information on the power of the input signal and moves the window moving means to a predetermined position based on the extracted information. As the transmission sub-codebook selecting means for obtaining the sub-codebook of, the first vector quantizing means and the second vector quantizing means perform the quantization based on a predetermined sub-codebook.

Alternatively, the dequantization means replaces the codebook composed of a plurality of sub-codebooks, and the first vector codeword and the first vector codeword corresponding to the first dequantization means adapted in advance to the characteristics of the output signal. One of the codebooks that accommodates a plurality of second vector codewords corresponding to the inverse vector quantization means of No. 2 and the ratio of the number of first vector codewords to the number of second vector codewords is changed. A window moving means for cutting out a part to be a sub-codebook is provided, and the reception sub-codebook selecting means extracts the information on the power of the transmitted input signal and sets the window moving means to a predetermined value based on the extracted information. The receiving sub-codebook selecting means for moving to a position to obtain a predetermined sub-codebook is used, and the first inverse vector quantizing means and the second inverse vector quantizing means perform dequantization based on the predetermined sub-codebook. I decided to do it.

Furthermore, the residual power is used as information regarding the power of the input signal, and a plurality of sub-codebooks are divided by the residual power to determine the ratio of the first vector codeword and the second vector codeword. I changed it to accommodate it.

Furthermore, a prediction error vector that uses a vector quantization means for quantizing a current input signal as the first vector quantization means and uses an input signal before a predetermined frame as the second vector quantization means It was used as a quantizer.

Furthermore, the first inverse vector quantizing means is an inverse vector quantizing means for inverse quantizing the present input signal, and the second inverse vector quantizing means is also an input signal before a predetermined frame. Inverse prediction error vector quantization means is used.

Further, the quantizing means stores, in place of the sub-codebook selecting means, the result of the quantization performed by the first vector quantizing means and the second vector quantizing means for each sub-codebook. Quantization distortion comparison means for comparison is provided, and the first vector quantization means and the second vector quantization means perform quantization for each sub-codebook and output to the quantization distortion comparison means. The quantizing distortion comparing means selects the sub-codebook which gives the minimum distortion among the distortion amounts for each sub-codebook, and transmits the number of the sub-codebook to the output corresponding part.

Further, the inverse quantizing means, in place of the sub-codebook selecting means, separates and extracts the sub-codebook number of the transmitted input signal, and specifies which of the sub-codebooks to select. Minimum distortion sub-codebook selecting means is provided, and the first inverse vector quantizing means selects the first vector code word by the designation in the input signal based on the sub-codebook selected based on the designation of the above selection. , 2nd inverse vector quantization means The second vector code word is selected.

The quantized dequantization system according to the present invention is provided with a transmission codebook in which a plurality of vector codewords matching the characteristics of the input signal are accommodated in advance, and the transmission codebook is provided for the first set area.
Of the first vector code word corresponding to the vector quantizing means and a plurality of transmission sub-codebooks in which the accommodation ratio of the second vector code word corresponding to the second vector quantizing means is changed. A transmission sub-codebook selecting means for configuring and extracting information regarding the power of the input signal and designating which one of the transmission sub-codebooks is selected, and the transmission sub-codebook selected based on the designation of the selection First vector quantizing means for selecting a suitable first vector codeword and second vector quantizing means for selecting a second vector codeword;
A vector code word that produces a small quantization distortion by comparing the first vector code word selected by the first vector quantization means and the second vector code word selected by the second vector quantization means. And a reception codebook containing a plurality of vector codewords matching the characteristics of the output signal in advance. , The number of words of the first vector codeword corresponding to the first inverse vector quantizing means and the second number corresponding to the second inverse vector quantizing means for the set area
The reception sub-codebook is constructed by a plurality of reception sub-codebooks having different accommodating ratios of vector codewords, the information on the power of the received input signal is extracted, and the reception sub-codebook is based on the extracted information. The first inverse codeword selecting means selects the first vector codeword according to the designation in the input signal by the receiving subcodebook selecting means for designating which of the two is selected and the receiving subcodebook selected based on the designation of the selection. Second inverse vector quantization means for selecting the vector quantization means and the second vector code word, the first vector code word selected by the first inverse vector quantization means, and the second inverse quantization. Second chosen by means
And a control means for controlling so as to select a decoding vector for which a small quantization distortion is obtained by comparing with the vector code word of.

Alternatively, instead of a transmission codebook consisting of a plurality of transmission subcodebooks, a first code that matches the characteristics of the input signal in advance is used.
Of the first vector code word corresponding to the vector quantizing means and a plurality of second vector code words corresponding to the second vector quantizing means, and the number of the first vector code word and the second vector code word And a transmission window moving means for cutting out a part of the codebook by changing the ratio of the number of vector codewords to be a transmission subcodebook, and the transmission subcodebook selecting means extracts information about the power of the input signal. A first vector quantizing means and a second vector quantizing means, which are transmission sub-codebook selecting means for obtaining a predetermined transmission sub-codebook by moving the transmission window moving means to a predetermined position based on the extracted information. Is a quantization means adapted to perform quantization based on the predetermined sub-codebook, and a reception codebook composed of a plurality of reception sub-codebooks,
A reception codebook that accommodates a plurality of first vector codewords corresponding to the first inverse vector quantization means and second vector codewords corresponding to the second inverse vector quantization means, which match the characteristics of the output signal in advance. , A reception window moving means for cutting out a part of the reception codebook and making it a reception subcodebook by changing the ratio of the number of words of the first vector codeword and the number of words of the second vector codeword. The selection means is a reception sub-codebook selection means for extracting information about the power of the transmitted input signal and moving the reception window moving means to a predetermined position based on the extracted information to obtain a predetermined reception sub-codebook. The first inverse vector quantizing means and the second inverse vector quantizing means are configured as an inverse quantizing device adapted to perform dequantization based on a predetermined reception sub-codebook.

[0031]

BEST MODE FOR CARRYING OUT THE INVENTION

Embodiment 1. Here, a quantizing device according to the present embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram of the configuration of a quantization device according to the first embodiment of the present invention. In the figure, a new configuration compared to FIG. 14 is a quantizer 1 having the configuration of the present embodiment, a residual power signal 6 obtained by linear prediction analysis of an input signal, and a scalar quantum of the residual power signal 6. Encoding means 7 for encoding and encoding, residual power for quantization 8, residual power code 11 for encoding the residual power for quantization, quantization residual power 8 output by the encoding means 7 and past Sub-codebook selection means 9 for selecting the codeword set A15 and the codeword set B16 using the delta residual power obtained from the quantized residual power of, and the subcodebook selection that is a subcodebook switching signal. The codebook 14 is a set of the signal 10 and the sub-codebook S _k (k = 1 ... K). Multiplexing means 2 for multiplexing the code number 22 output by the comparing means 21 and the quantized residual power code 11 output by the encoding means 7 and sending out to the transmission line as the transmission line code 24.
3 is outside the quantizer and is conventionally necessary as an element for multiplexing transmission of a quantized codeword of a voice vector and transmission of other voice information. The same parts as those in FIG. 14 are designated by the same reference numerals and the description thereof will be omitted. Although the encoding means 7 is provided in the quantizer for the sake of convenience in the above description, the power information may be obtained from it because it needs to be transmitted elsewhere.

The operation will be described below centering on the above-mentioned new structure. First, in the quantizer 1, for example, 8 kHz
The digital audio signal sampled by z is input as the input signal 3. In the input signal analysis means 4, the input signal 3
For example, the linear prediction analysis is performed for each 20 ms frame length to obtain the n-th order linear prediction parameter that is the spectral envelope information of the voice, and then converted into, for example, a line spectrum pair (LSP), and an n-dimensional input is performed. The vector 5 is output to the vector quantization unit 12 and the prediction error vector quantization unit 13. At the same time, for example, the residual power signal 6 obtained from the linear prediction analysis result is obtained as an essential transmission parameter other than the spectrum envelope information, and is output to the encoding means 7.

In the encoding means 7, the residual power signal 6 is
For example, scalar quantization is performed with a resolution of 6 bits, and the obtained result is transmitted as a quantized residual power 8 to a sub-codebook selecting unit 9 that selects a sub-codebook, and then encoded to obtain a residual power code. It is sent as 11 to the multiplexing means 23. The sub-codebook selecting unit 9 has a storage unit such as a memory for storing the past quantized residual power, and the quantization residual power 8 and the past quantized residual stored in the memory. Taking the difference from the power, for example, Δ
Get the residual power. The sub codebook selecting means 9 obtains the obtained Δ.
The sub codebook selection signal 10 is output based on the residual power, and the sub codebook selection switches 48 and 49 are controlled to select the sub codebook S _k from the codebook 14. The sub codebook selection switches 48 and 49 are interlocked.

For each sub-codebook S _k , for example,
The codeword set A15 is composed of vector code words of positive integer M _k , and the code word set B16 is composed of N−M _k prediction error vector code words, that is, the sub codebook is N.
It consists of a number of vector code words. This is supplied to the vector quantization means 12 and the prediction error vector quantization means 13 as a code word candidate.

The input vector 5 is the vector quantization means 1
2 and the prediction error vector quantizer 13 are input.
The vector quantizer 12 calculates, for example, the Euclidean distance between the vector codeword in the codeword set A15 in the subcodebook S _k selected by the subcodebook selector 9 from the codebook 14 and the input vector X. Code number 1 of the optimum vector code word that is calculated and gives the smallest distance among them
7 and the minimum Euclidean distance at that time is the quantization distortion 18
Is output to the comparison means 21.

The prediction error vector quantization means 1
3 is the past code number 22 transmitted up to the current processing time point.
A prediction value X _p of the current input vector 5 is obtained by using the quantization result of the past input vector represented by, and the prediction error vector X between the prediction value X _p and the input vector X is calculated.
Calculate _e . Subsequently, similarly to the calculation of the distance of the vector codeword, the codeword set B in the sub codebook S _k is calculated.
The Euclidean distance between the prediction error vector codeword in 16 and the prediction error vector X _e is calculated, and the code number 19 of the optimum prediction error vector codeword that gives the smallest distance among them and the smallest Euclidean value at that time The distance is output to the comparison means 21 as the quantization distortion 20.

The comparing means 21 compares the vector quantizing means 12 with the two quantizing distortions 18 and 20 inputted from the predictive error vector quantizing means 13 and gives a code number 22 giving a small quantizing distortion. Is output to the prediction error vector quantization unit 13 as the past input vector quantization result, and the code number 22 output from the comparison unit 21 and the residual power code output from the encoding unit 7 are output. 11 and 11 are multiplexed by the multiplexing means 23, and are transmitted to the transmission line as the transmission line signal 24. In this way, when the vector quantization of the input signal, the current value is quantized,
Since two kinds of quantization are performed to quantize a value predicted from a temporally previous value, and both are compared, quantization with high accuracy can be performed.

FIG. 2 is a diagram for explaining the operation of the quantizer of FIG. As a whole, this will be described later with respect to the LSP and the absolute value of the Δ residual power when the vector codeword based on the current value is selected when the quantizer having a single fixed codeword ratio is operated. 3 shows a distribution of the correlation value lsp _c obtained by the equation (1). FIG.
Is a single codebook as a whole, when a quantizer with a single fixed codeword ratio is operated, as in the case where a prediction error vector codeword that also combines past values is selected,
It is a distribution of the correlation value lsp _c with the absolute value of the Δ residual power. The horizontal axis is Δ residual power (unit is d
B). Also, the vertical axis represents the LSP of the immediately preceding frame.
And the correlation value lsp _c between the LSPs of the current frame are shown, and the closer the correlation value is to 1, the more stationary the voice. 2 and FIG. 3, the probability that the vector codeword of FIG. 2 is selected is high in a portion where the absolute value of Δ residual power is large, and conversely, the absolute value of Δ residual power is small. In part, there is a high probability that the prediction error vector codeword of FIG. 3 will be selected. When the Δ residual power has a positive value, for example, it indicates a rising portion of a voice that shifts from a consonant to a vowel, and when it has a negative value, for example, a voice that shifts from a stationary vowel unit to a silent section is used. The falling part is shown.

Further, according to FIG. 2 and FIG. 3, the frame having a very high inter-frame correlation of the LSP having the correlation value lsp _c of 0.998 or more is a voice stationary part such as a vowel or some consonants. In these frames, the prediction error vector codeword in FIG. 3 has a high probability of being selected, and in the frames in which the correlation value is less than 0.998, the vector codeword in FIG. 2 has a high probability of being selected. Further, according to FIG. 3, when the prediction error vector codeword is selected, the absolute value of the Δ residual power is 0.8 dB or less, and the Δ
When the absolute value of the residual power is larger than 0.8 dB,
Only vector codewords are selected.

Further, x in the following equation (1) is the LSP
It represents an n-dimensional vector of coefficients, and i represents each order of the LSP. Further, t corresponds to the analysis frame number of the input voice vector, for example.

[0041]

[Equation 1]

From the LSP correlation performance charts shown in FIGS. 2 and 3, for example, if the absolute value of Δ residual power and the positive / negative sign are used as the codebook switching signal, the vector codeword and the prediction error vector codeword are obtained. It is possible to selectively specialize the codebook structure, and as a result, it is possible to improve the quantization efficiency. Sub-codebook S _k (k =
1. . . The number K of K) is, for example, as shown in FIG.
The sign of the Δ residual power is judged according to the sign and the thresholds α and β, and classified into four. The size of each codebook is, for example, the total number of vector codewords N = 256 (8 bits). Α and β in the figure are thresholds of the absolute value of the Δ residual power, and here, α = β = 0.8 dB.

As a method of creating the sub-codebook S _k (k = 1 ... K), for example, learning data is input to the quantizer 1 of FIG. 1 instead of the input signal 3, and the input signal analysis means 4, At the same time that the encoding means 7 is operated, for example, the sub-codebook selecting means 35 is operated in accordance with the classification rule of FIG. 6, and the output input vector 5 is output in accordance with the sub-codebook selecting signal 10 for each sub-codebook S. Learning data is classified by _k (k = 1 ... K), and the learning data for each sub-codebook is
For example, it is possible to employ a method in which learning is performed by clustering each independently by the LBG method. Regarding the algorithm of the LBG method, Y. Linde et al., “An Algorthm for Vect
or Quantizer Design ”(IEEE Transactions on Commun
ications Vol. COM-28,1, pp.84-95, Jan 1980), detailed explanation is omitted.

As a result of performing the learning of the sub-codebook based on the Δ residual power as described above, each sub-codebook is a code word specialized for a spectral feature that differs depending on the magnitude of the Δ residual power. Will be composed. For example, the sub codebook S
In 4, the frame with the Δ residual power larger than α is selected, so that the LSP correlation value is small and the part showing the change in the rising of the voice, that is, the voice transition part frame rising from the consonant part such as the head to the vowel Include a vector codeword belonging to the vector codeword and a prediction error vector codeword, and in the sub-codebook S2, since a frame with Δ residual power larger than 0 and smaller than α is selected, the LSP correlation value is large. , And a frame showing a rising change, that is,
It includes a vector code word belonging to a frame in which a spectrum change is gentle such as a transition portion from a vowel transition portion to a vowel steady portion and a vowel steady portion, and a prediction error vector code word. As a result, the quantizing distortions 18 and 20 input to the comparing means 21 in the present embodiment are reduced as compared with the quantizing distortion of the conventional single-codebook-type quantizing device. In addition,
In FIG. 4, the LSP coefficient obtained from the input voice is the above-mentioned Δ.
The elements classified according to the sub-codebook are shown according to the rule regarding the residual power.

For example, as shown in FIG. 5, the Δ residual power is high as a dynamic scale of the cepstrum, which is often used for speech labeling, as an index for efficiently extracting a stationary portion or a transient portion of speech. Show correlation. Regarding the physical meaning of the dynamic scale of the cepstrum and its derivation method, Sagayama et al., “Individuality information included in the dynamic scale of speech” (Proceedings of the Acoustical Society of Japan, 3-2-7, 1979). The detailed description is given in, so the description is omitted.

The above is the outline of the quantizer according to the first embodiment. According to the first embodiment, it is possible to select a codebook for quantization according to the characteristics of the input signal, so that it is possible to perform quantization efficiently and improve the quality of the quantization device. . That is, as the predetermined parameter,
Since the Δ residual power, which has a high correlation with the dynamic scale of speech, is used, for example, it is possible to particularly reduce the quantization distortion of the frame in which the phonological feature changes from a vowel important for speech perception to another vowel. It is possible to configure each codebook and perform quantization specialized for an input signal.

According to the first embodiment, the residual power information used for selecting the sub codebook is an essential transmission parameter and is separately multiplexed and transmitted. Therefore, the quantization code is specially used. Since it is not necessary to transmit the sub codebook selection flag and the amount of information is not increased, the quantization transmission efficiency is increased.

In the first embodiment, the LBG method is used as the codebook learning method, but of course, as another method, for example, the k-means method is used to perform the codebook learning. The effect of the invention does not change. The sub-codebook classification thresholds α and β described above can of course take arbitrary values, and the number of thresholds may be two or more.

Embodiment 2 Next, the dequantization device of the present invention will be described with reference to the drawings. FIG. 7 is a configuration block diagram of the inverse quantization device according to the second embodiment. In the figure, the configuration newly added from FIG. 15 is the quantization residual power 26 and the quantization residual power 26 obtained by the separating means 25 described later in the dequantization device 2 having the configuration of this embodiment. In response, the Δ residual power is decoded and the sub codebook selection means 28 for outputting the sub codebook selection signal 29, the sub codebook selection signal 29 for switching the sub codebook S _k ,
Codebook 33, codeword set A34, codeword set B3
5, sub codebook selection switches 48 and 49. The same parts as those in FIG. 15 are designated by the same reference numerals and the description thereof will be omitted. Further, the code number 27 for designating the vector code word or the prediction error vector code word and the residual power code are separated from the transmission line signal 24, and further, the separating means 25 for decoding the quantized residual power 26 is inverse quantization Conventionally, it is necessary as an element outside the device for receiving voice information and separating the received signal into a voice vector to be dequantized and other information. Therefore, in this figure, although described as an element for dequantization for convenience, the residual power signal is originally obtained from the multiplexed and transmitted information.

The operation will be described below centering on the above new structure. First, the separating means 25 receiving the transmission path signal 24 from the transmission path separates the code number 27 and the residual power code and then decodes the quantized residual power 26. It is output to the codebook selecting means 28.

The sub-codebook selecting means 28 has a storage means such as a memory, for example, and the quantization residual power 26 output from the separating means 25 and the past quantization residual power held in the memory. The quantized Δ residual power is decoded by taking the difference between the sub-codebook selection signal 29 and the sub-codebook selection signal 29.
8 and 49 are switched to select the corresponding sub codebook S _k from the codebook 33. The sub codebook selection switch 48,
49 works together.

When the code number 27 output by the separating means 25 designates a code word in the code word set A34 in the sub codebook S _k , the control means 30 causes the inverse vector quantization means 31 to perform the code. The number 27 is output and the sub codebook S is output.
_When designating a codeword in the codeword set B35 in _k, the code number 27 is output to the inverse prediction error vector quantization means 32, and the changeover switch 50 is operated by the control signal 36 to perform output control. .

The inverse vector quantizing means 31 reads the vector code word in the code word set A34 in the sub codebook S _k designated by the input code number 27, and outputs the decoded vector. The inverse prediction error vector quantization means 32,
The prediction error vector codeword in the codeword set B35 in the sub-codebook S _k specified by the input code number 27 is read, and the prediction value of the current input vector is obtained using the past decoded vector 37. Then, the prediction value of the input vector is added to the prediction error vector code word to generate a decoded vector, which is output as the decoded vector 38 via the changeover switch 50.

The above is the outline of the inverse quantization apparatus according to the second embodiment. According to the second embodiment, the codebook for dequantization can be selected according to the characteristics of the input signal, so that dequantization can be efficiently performed and the quality of the dequantization device is improved. be able to. That is, since the Δ residual power, which has a high correlation with the dynamic scale of speech, is used as the predetermined parameter, for example, the quantization distortion of the frame in which the phonological feature changes from a vowel important for speech perception to another vowel. The sub-codebooks can be configured so as to particularly reduce, and dequantization specialized for the input signal can be performed.

Further, according to the second embodiment, since the residual power information used for selecting the sub codebook is a transmission parameter that is separately required, the sub codebook selection flag is specially transmitted and received. It is possible to improve the transmission efficiency for inverse quantization without increasing the amount of information.

In the second embodiment, the codebook and sub-codebook selecting means in the inverse quantizer shown in FIG. 7 are not limited to the same ones as the quantizer on the transmitting side, but of course. The inverse quantizing process may be performed using the same codebook and sub-codebook selecting means of the quantizing device shown in FIG.

Embodiment 3 According to FIG. 2, a normal vector codeword is easily selected in a portion where the absolute value of the Δ residual power is large, and according to FIG. 3, in a portion where the absolute value of the Δ residual power is small, The prediction error vector codeword is easy to select. In this embodiment, focusing on this,
The sub-codebook S belonging to the part where the Δ residual power is large.
_k (k = 1,4) is configured such that the number of vector code words is larger than that of the prediction error vector code words, and
Sub-codebook S _k belonging to a portion where the absolute value of the residual power is small
On the other hand, (k = 2, 3) is configured such that the number of prediction error vector code words is larger than that of vector code words. Hereinafter, the operation will be described focusing on the above new configuration. FIG. 8 is a block configuration diagram of the quantization device according to the present embodiment.
In the figure, new elements are a codebook 39, a codeword set A40, and a codeword set B41. The other elements are the same as those in FIG. Further, FIG. 9 is a configuration diagram of the codebook in FIG.

As the structure of each vector code word of the sub codebook in this embodiment, for example, as in the case of FIG. 6B, the division threshold α is set to 0.8 dB, and the absolute value of the Δ residual power is set. When the value is smaller than 0.8 dB (subcodebooks S ₂ and S _{3 in} FIG. 9), for example, the number of vector codewords that are codeword set A is 64, and the prediction error that is codeword set B is When the number of vector code words is 192, the ratio of prediction error vector code words is increased, and when the absolute value of the Δ residual power is larger than 0.8 dB (sub-codebook S ₁ in FIG. 9, S ₄ ), the number of vector codewords is 192,
The number of prediction error vector code words is set to 64, and the ratio of vector code words is increased. Further, FIG. 10 shows that the codebook of the quantizing device according to the present embodiment is extracted to obtain the codebook 3
FIG. 9 is a detailed diagram for easily explaining the configurations of vector code words and prediction error vector code words inside the sub codebook S _k (k = 1 ... 4) in 9 and how they are selected. Α in the figure is a threshold value used for codebook classification.

When the sub-codebook S ₁ or S ₄ is selected, the vector quantizing means uses the code numbers 0 to 191.
No. up to No. 221 are searched as vector code word candidates, and the prediction error vector quantization means searches from No. 192 to No. 255 as prediction error vector code word candidates. When the sub-codebook S ₂ or S ₃ is selected, the vector quantization means searches the code numbers 0 to 63 as vector code word candidates, and the prediction error vector quantization means 6
Numbers 4 to 255 are searched as prediction error vector codeword candidates. The sub-codebook selecting means 28 inspects the positive / negative sign and absolute value of the Δ residual power according to the classification thresholds α and β set above and the sub-codebook classification rule of FIG. The codebook S _k is selected.

The above is the outline of the quantizer according to the third embodiment. According to the present embodiment, the fact that the Δ residual power is small in the stationary part of the voice and the Δ residual power is large in the transient part of the voice is used, and the sub codebook is used in the place where the absolute value is large. The S, S ₄ vector codewords are increased and the prediction error vector codewords are reduced, and the ratio is changed to the opposite ratio. The important part can be quantized with high precision.

Embodiment 4 It is also possible to apply the codebook in the above-described Embodiment 3 to the dequantization apparatus in Embodiment 2 described above. For example, FIG. 11 is a diagram in which the codebook 33 of the dequantizer of the second embodiment is replaced with the codebook 42 of the third embodiment. According to the fourth embodiment, the fact that the Δ residual power is small in the stationary part of the voice and the Δ residual power is large in the transient part of the voice is used, and the sub-codebook is used depending on the magnitude of the absolute value. Since the vector codewords of (1) are increased / decreased and the number of words of the prediction error vector codewords is reversed, it is possible to accurately dequantize the parts that are important for speech perception, such as the voice stationary part and the voice transient part.

Embodiment 5 In each of the above embodiments, the codebook is composed of a plurality of sub-codebooks (for example, S ₁ to S ₄ ), and information about power (for example, Δ residual power)
The device having a configuration for designating which sub-codebook to switch and select has been described above. In the present embodiment, a large single area is prepared as a codebook, in which one vector codeword is sequentially arranged and the other prediction error vector codeword is sequentially arranged. For example, the area is assigned address 384, and the code numbers 0 to 191 of the vector code word are assigned to the addresses in the first half. Code numbers 0 to 191 of the prediction error vector code word are assigned to the latter half addresses.

In the above codebook, the size of the subcodebook is, for example, 25 due to restrictions on transmission or quantization processing.
Since there is a constraint of a width of 6 words, a window is set and a sub-codebook of 256 words is cut out from the codebook of 384 words by a window.
The sub-codebook selecting means moves the above-mentioned window by the Δ residual power, and when selecting the sub-codebooks S ₂ and S ₃ , the prediction error vector codeword is cut out in large numbers, and the sub-codebook S is selected. _When selecting ₁ and S ₄ , the vector codeword is moved so as to be extracted. Subsequent operations are the same as those described in the first to fourth embodiments. Of course, both the quantizer and the dequantizer can be applied.

Embodiment 6 FIG. As input parameters of the sub-codebook selecting means of the first to fifth embodiments, in addition to the Δ residual power, for example, pitch and code-driven linear predictive coding (C
A plurality of pieces of information such as long-term prediction such as a flag of an adaptive codebook in the ELP) system and voice / unvoiced information may be used together. That is, by using the long-term prediction of pitch or the like, it is possible to more accurately extract a voice signal stationary portion such as a vowel, and it is possible to perform accurate quantization. Also,
By using the voice / unvoiced information, for example, the sub-codebook can be switched between the vowel part and the consonant part, so that correct quantization can be performed.

Further, in each of the embodiments, the case where the Δ residual power which is a parameter representing the power contained in the input signal is used as the sub-codebook selecting means has been described. In the present embodiment, as the information regarding the power, for example, the same effect can be obtained by using the frame energy and its difference value or the frequency band division of the frame using a bandpass filter, and using the obtained band-based frame energy and their difference value. Have. Alternatively, the Δ residual power and any one of the above values may be added and used, and the corresponding sub-code may also be used based on the distinction between them.

Furthermore, instead of quantizing and transmitting the residual power, for example, the above-described cepstrum dynamic measure may be scalar-quantized and transmitted.

In the apparatus of each of the above embodiments, the codebook uses codewords corresponding to different quantization or dequantization of codeword set A and codeword set B. For example, focusing on quantization, different quantization is
Specifically, they are vector quantization and prediction error vector quantization. The first and second different quantizations are:
In addition to vector quantization and prediction error vector quantization, matrix quantization, multi-stage vector quantization, split (sp
lit) There may be quantization and the like. Any of the above quantizations may be used as the first quantization corresponding to the vector quantization in the above embodiment and the second quantization corresponding to the prediction error vector quantization. The codebook is of course a codebook corresponding to it.

Further, in the above-mentioned embodiment, the two quantizing means of the first quantizing and the second quantizing are used, and the comparing means selects one having less distortion from the two selected codewords. However, the above-mentioned other quantizing means are also used together, for example, three different quantizing means and corresponding codebooks are used,
The comparing means may select the code word with the least distortion among the three selected code words.

Embodiment 7 In the fourth embodiment, the constituent ratios of the codeword sets A and B are the same between the sub-codebooks 2 and 3 and between the sub-codebooks 1 and 4, but of course the codewords of the sub-codebook are the same. The composition ratio of the set can be arbitrarily set separately in each of the sub codebooks S ₁ , S ₂ , S ₃ , and S ₄ . The codebook classification threshold value α of the sub-codebook in the above embodiment can be set at the time of learning the codebook. In this way, the sub-codebook can be classified according to the codebook learning data by using the threshold value of the sub-codebook classification that is optimal for the data used for codebook learning. And an inverse quantizer is obtained.

Embodiment 8 FIG. In the first to fourth embodiments, the sub-codebook is selected in place of the sub-codebook selecting means by the quantization distortion minimization criterion, and the sub-codebook number is transmitted instead of the essential transmission parameter. The sub codebook may be selected. This is effective when, for some reason, an essential transmission parameter such as residual power cannot be obtained. In this case, the amount of transmission information cannot be reduced, but there is an advantage that a plurality of sub-codebooks are selected and two quantizations are further selected to transmit a high-quality quantized code. A quantizer according to the present invention will be described with reference to the drawings. FIG. 12 is a block diagram showing the configuration of the eighth embodiment of the present invention. In the figure, a configuration newly added as compared with FIG. 1 is a quantization distortion comparing means 45 and a sub codebook number 46. Other elements are shown in FIG.
Therefore, the description is omitted.

The quantizing distortion comparing means 45 outputs the sub-codebook selection signal 10 instead of the sub-codebook selecting means 9 of FIG. 1, and sequentially outputs the sub-codebook S _k (k = 1 ... K). The sub codebook selection switches 48 and 49 are switched so as to scan. The vector quantization means 12 and the prediction error vector means 13 are
According to the sub-codebook selection signal 10, all sub-codebooks S _k
For example, the Euclidean distance of the input vector 5 is obtained as the quantization distortions 18 and 20 for the code word of (k = 1 ... K). For example, if k = 4, S ₁ to S ₄ are obtained, and a total of 4
Code number 1 that gives the least quantization distortion to the time vector code word
Output as 7 and 19. This output and the quantization distortion 1
8 and 20 are sequentially sent to the quantization distortion comparing means 45.

[0072] Furthermore, quantization distortion comparison section 45, the sub-codebook S _k as the quantization distortion minimum by comparing the quantization distortion 18 of each sub-codebook _{S k (k = 1 ... K} ) Is selected, the sub codebook number 46 is sent to the multiplexing means 23, and the code numbers 17 and 19 are sent to the comparing means 21. The comparing means 21 selects one of the code numbers 17 and 19 with the smaller quantization distortion and outputs it as the code number 22 to the multiplexing means 23. The multiplexing means 23 multiplexes the code number 22 and the sub-codebook number 46 and sends them as a transmission line signal 24 to the transmission line.

The above is the outline of the quantizer according to the eighth embodiment. With the configuration of the eighth embodiment, in addition to performing quantization with a codebook specialized for a characteristic portion important for speech perception, quantization can also be performed with a quantization distortion minimization criterion.

Ninth Embodiment An inverse quantizer corresponding to the eighth embodiment will be described. Using the transmitted sub-codebook number 47 instead of the predetermined parameter, the sub-codebook that minimizes the quantization distortion is selected and the inverse quantization is performed. The dequantization device of this embodiment will be described with reference to the block diagram of FIG. In the figure, the components newly added as compared with FIG. 7 are a sub-codebook number 47 and a minimum distortion sub-codebook selecting means 51. The other elements are the same as those in FIG. 7, and the description thereof will be omitted.

The separating means 25 receiving the transmission path signal 24,
The transmission line signal 24 is assigned a code number 27 and a sub-codebook number 47.
To separate. The minimum distortion sub-codebook selection means 51 receives the sub-codebook number 47, outputs the sub-codebook selection signal 29, switches the sub-codebook selection switches 48 and 49, and selects the corresponding sub-codebook S _k . . The sub codebook selection switch 4
8,49 work together.

When the code number 27 output by the separating means 25 designates a code word in the code word set A34 in the sub codebook S _k , the control means 30 causes the inverse vector quantization means 31 to perform the code. The number 27 is output and the sub codebook S is output.
_When designating a codeword in the codeword set B35 in _k, the code number 27 is output to the inverse prediction error vector quantization means 32, and the changeover switch 50 is operated by the control signal 36 to perform output control. .

The inverse vector quantization means 31 reads the vector code word in the code word set A34 in the sub codebook S _k designated by the input code number 27, and outputs the decoded vector. The inverse prediction error vector quantization means 32,
The prediction error vector codeword in the codeword set B35 in the sub-codebook S _k specified by the input code number 27 is read, and the prediction value of the current input vector is obtained using the past decoded vector 37. Then, the prediction value of the input vector is added to the prediction error vector code word to generate a decoded vector, which is output as the decoded vector 38 via the changeover switch 50.

The above is the outline of the quantizer according to the ninth embodiment. With the configuration of this embodiment, it is possible to perform dequantization with a codebook specialized for a characteristic part important for speech perception, and also to perform dequantization with a quantization distortion minimization criterion.

In the ninth embodiment, dequantization is performed by the codebook in the dequantization device, which does not necessarily match the codebook on the transmission side shown in FIG. 13, but the quantum shown in FIG. The inverse quantization process may be performed using the same codebook as the coding device.

[0080]

As described above, according to the present invention, a codebook having a plurality of subcodebooks having different vector codewords,
Since the transmission sub-codebook selection means, the first and second vector quantization means, and the comparison means are provided, there is an effect that high-quality vector quantization can be performed without increasing the information transmission amount.

Since the codebook includes a plurality of subcodebooks having different vector codewords, the reception subcodebook selecting means, the first and second inverse vector quantizing means, and the control means. There is an effect that high quality inverse vector quantization can be performed with a small amount of information transmission.

Since a large codebook and window moving means for cutting out different vector codewords at different ratios are provided and the transmission sub-codebook selecting means moves the window to a predetermined position, the amount of information transmission is increased. There is an effect that high-quality vector quantization can be performed without using it.

Since a large codebook and window moving means for cutting out different codewords at different ratios are provided, and the receiving sub-codebook selecting means moves the window to a predetermined position,
There is an effect that high quality inverse vector quantization can be performed with a small amount of information transmission.

Furthermore, since the residual power is used as the information regarding the power of the input signal, in addition to the basic effect, there is an effect that quantization suitable for the characteristics of the voice can be performed.

Furthermore, since the vector quantizing means and the prediction error vector quantizing means quantize and the comparing means selects the quantized data, it is possible to quantize in accordance with the characteristics of the voice with a relatively simple structure. There is an effect that can be done.

Furthermore, since the inverse vector quantizing means and the inverse prediction error vector quantizing means perform inverse quantization and the control means selects the decoding vector, the characteristics of the voice are matched with a relatively simple structure. There is an effect that inverse quantization can be performed.

Further, since the quantizing distortion comparing means is provided instead of the sub codebook selecting means, there is an effect that high quality vector quantizing with minimum distortion can be performed.

Furthermore, since the minimum distortion sub-codebook selecting means is provided instead of the sub-codebook selecting means, there is an effect that high-quality inverse vector quantization with minimum distortion can be performed.

Further, a transmission codebook composed of a plurality of subcodebooks containing different types of vector codewords at different ratios, a transmission subcodebook selecting means, a first vector quantizing means, and a second vector quantizing means. Inverse quantum including a quantizer including a comparing unit, a reception codebook including a plurality of subcodebooks, a reception subcodebook selecting unit, a first inverse vector quantizing unit, and a second inverse vector quantizing unit. Since it is configured with an encoding device, there is an effect that high-quality vector quantization and inverse vector quantization can be performed without increasing the amount of information transmission.

Quantization comprising a large transmission codebook and a transmission codebook selecting means, a first vector quantizing means and a second vector quantizing means for instructing to cut out a part of the large transmission codebook and use it as a transmission subcodebook. Means, a reception sub-codebook selecting means for instructing to cut out a large reception codebook and a part thereof, an inverse quantizer including a first inverse vector quantizing means and a second inverse vector quantizing means. Therefore, there is an effect that high quality vector quantization and inverse vector quantization can be performed without increasing the amount of information transmission.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a quantization device according to a first embodiment of the present invention.

FIG. 2 is a diagram showing a correlation distribution between a Δ residual power and a previous-frame / current-frame correlation function lsp _c when a vector codeword is selected with a single codeword ratio in a single codebook.

FIG. 3 is a diagram showing a correlation distribution between Δ residual power and a correlation function lsp _c between a previous frame and a current frame when a prediction error vector codeword is selected with a single codeword ratio in a single codebook. is there.

FIG. 4 is a diagram showing how LSP coefficients are classified by sub-codebook according to Δ residual power.

FIG. 5 is a diagram showing a correlation between a voice signal, a cepstrum dynamic scale, and a Δ residual power.

FIG. 6 is a diagram showing classification of a sub codebook according to Δ residual power.

FIG. 7 is a configuration block diagram of an inverse quantization device according to the second embodiment of the present invention.

FIG. 8 is a configuration block diagram of a quantization device according to a third embodiment of the present invention.

FIG. 9 is a diagram showing a configuration example of a codebook according to the third embodiment.

FIG. 10 is a diagram showing details of a codebook and a codeword selection relationship according to the third embodiment.

FIG. 11 is a block diagram of a configuration of an inverse quantization device according to a fourth embodiment of the present invention.

FIG. 12 is a block diagram of a quantizer in Embodiment 8 of the present invention.

FIG. 13 is a configuration block diagram of an inverse quantization device according to a ninth embodiment of the present invention.

FIG. 14 is a configuration block diagram of a first conventional quantization device.

FIG. 15 is a configuration block diagram of a first conventional inverse quantization device.

[Fig. 16] Fig. 16 is a configuration block diagram of a second conventional encoding device including a plurality of codebooks.

17 is a configuration block diagram of a decoding device corresponding to the encoding device of FIG.

[Explanation of symbols]

1 quantizer, 2 inverse quantizer, 3 input signals, 4
Input signal analyzing means, 5 input vectors, 6 residual power signal, 7 encoding means, 8,26 quantized residual power, 9,28 sub-codebook selecting means, 10,29 sub-codebook selecting signal, 11 residual Power code, 12 vector quantization means, 13 prediction error vector quantization means, 14, 3
3, 39, 42 Codebook, 15, 34, 40, 43 Codeword set A, 16, 35, 41, 44 Codeword set B, 17, 19, 27 Code number, 18, 20 Quantization distortion, 21 Comparison means , 22 past code number, 23 multiplexing means, 24 transmission line signal, 25 demultiplexing means, 30
Control means, 31 inverse vector quantization means, 32 inverse prediction error vector quantization means, 36 control signal, 37 past decoded vector, 38 decoded vector, 45 quantization distortion comparison means, 46, 47 sub codebook number, 48, 49 sub codebook selection switch, 50 changeover switch, 51 minimum distortion sub codebook selection means.

Claims (11)

[Claims] 1. A codebook in which a plurality of vector codewords adapted to the characteristics of an input signal are accommodated in advance, and the number of first vector codewords corresponding to a first vector quantizing means, which will be described later, is set for a set area and The codebook is configured by a plurality of sub-codebooks in which the accommodation ratio of the second vector codeword corresponding to the second vector quantization means is changed, information on the power of the input signal is extracted, and the extracted information is extracted. Transmission sub-codebook selecting means for designating which one of the plurality of sub-codebooks to select based on the above, and a first vector codeword adapted to the input signal by the sub-codebook selected based on the designation of the selection. A first vector quantizing means for selecting, a second vector quantizing means for selecting a second vector codeword, and a first vector codeword selected by the first vector quantizing means, Second vector quantity The second selected by the child-bearing means
And a comparison means for transmitting a selection result to an output equivalent part for outputting information regarding the power of the input signal. apparatus. 2. A codebook in which a plurality of vector codewords adapted to the characteristics of an output signal are accommodated in advance, and the number of first vector codewords corresponding to a first inverse vector quantizing means, which will be described later, for a set area. And the second inverse vector quantizer corresponding to the second
The above-mentioned codebook is configured with a plurality of subcodebooks having different vector code word accommodation ratios, information on the power of the received input signal is extracted, and one of the subcodebooks is selected based on the extracted information. The first inverse vector quantization for selecting the first vector code word by the designation in the input signal by the receiving sub-codebook selecting means for designating whether or not to carry out, and the sub-codebook selected based on the designation of the selection. Means and a second inverse vector quantization means for selecting a second vector code word, a first vector code word selected by the first inverse vector quantization means, and a second inverse vector quantization And a control means for controlling so as to select a decoding vector that gives a small quantization distortion by comparing with the second vector codeword selected by the means. 3. A codebook consisting of a plurality of sub-codebooks, and a first vector codeword corresponding to a first vector quantization means adapted to a characteristic of an input signal in advance and a second vector quantization means supported. Of the second vector codeword, and a part of the codebook is cut out by changing the ratio of the number of words of the first vector codeword to the number of words of the second vector codeword. The transmission sub-codebook selecting means extracts information about the power of the input signal, and moves the window moving means to a predetermined position based on the extracted information. Transmission sub-codebook selecting means for obtaining a sub-codebook, wherein the first vector quantizing means and the second vector quantizing means perform quantization based on the predetermined sub-codebook. The quantizing device according to claim 1. 4. A first vector codeword and a second inverse vector quantization corresponding to a first inverse vector quantization means adapted to a characteristic of an output signal in advance, instead of a codebook including a plurality of sub-codebooks. A codebook containing a plurality of second vector codewords corresponding to the means, the number of words of the first vector codeword, and the second
A window moving means for cutting out a part of the codebook by changing the ratio of the number of vector codewords to a sub-codebook is provided, and the reception sub-codebook selecting means provides information on the power of the transmitted input signal. The receiving sub-codebook selecting means for extracting and moving the window moving means to a predetermined position based on the extracted information to obtain a predetermined sub-codebook, the first inverse vector quantizing means and the second inverse vector The dequantization device according to claim 2, wherein the quantization means performs dequantization based on the predetermined sub-codebook. 5. The residual power is used as the information on the power of the input signal, and a plurality of sub-codebooks are divided by the residual power to change the ratio of the first vector codeword and the second vector codeword. The quantizing device according to claim 1 or 3, or the dequantizing device according to claim 2 or 4, characterized in that the quantizing device is accommodated as a unit. 6. A prediction error vector quantum, wherein the first vector quantizing means is a vector quantizing means for quantizing a current input signal, and the second vector quantizing means is also using an input signal before a predetermined frame. 4. The quantizing device according to claim 1, wherein the quantizing device is a quantizing device. 7. An inverse vector quantizing means for inverse quantizing the current input signal as the first inverse vector quantizing means,
5. The inverse quantization device according to claim 2, wherein the second inverse vector quantization means is an inverse prediction error vector quantization means that also uses an input signal before a predetermined frame. 8. Quantization distortion for storing and comparing results of quantization performed by the first vector quantizing means and the second vector quantizing means for each sub-codebook instead of the sub-codebook selecting means. Comparing means is provided, and the first vector quantizing means and the second vector quantizing means perform quantization for each sub-codebook and output it to the quantizing distortion comparing means. The means for selecting a sub-codebook which gives the minimum distortion among the distortion amounts for each sub-codebook, and transmitting the number of the sub-codebook to the output corresponding portion. Quantizer. 9. A minimum-distortion sub-codebook selection designating which sub-codebook is selected by separating and extracting the sub-codebook number of the transmitted input signal in place of the sub-codebook selecting means. Means is provided, based on the sub-codebook selected based on the designation of the above selection,
The first inverse vector quantization means selects the first vector code word and the second inverse vector quantization means selects the second vector code word according to the designation in the input signal. The dequantization device according to claim 2. 10. A transmission codebook in which a plurality of vector codewords matching the characteristics of an input signal are accommodated in advance, and the number of first vector codewords corresponding to the first vector quantization means and the The transmission sub-codebook is composed of a plurality of transmission sub-codebooks in which the accommodation ratio of the second vector codeword corresponding to the second vector quantizer is changed, and the transmission subcodebook is extracted by extracting the information on the power of the input signal. A transmission sub-codebook selecting means for designating which one of the codebooks is selected, and a first vector codeword adapted for an input signal by the transmission sub-codebook selected based on the designation of the selection. Vector quantizing means, second vector quantizing means for selecting the second vector code word, and first vector code word and second vector quantizing means selected by the first vector quantizing means. No. selected in A quantizer having a comparison means for comparing the two vector codewords with each other to select a vector codeword for which a small quantization distortion is obtained and transmitting the selected information to the output corresponding portion for outputting information on the power of the input signal; In a reception codebook containing a plurality of vector codewords that match the characteristics of the output signal in advance, the number of first vector codewords corresponding to the first inverse vector quantization means and the second inverse vector for the set area. The reception codebook is composed of a plurality of reception sub-codebooks in which the accommodating ratio of the second vector codeword corresponding to the quantizing means is changed, and information on the power of the received input signal is extracted and extracted. The receiving sub-codebook selecting means for designating which one of the receiving sub-codebooks is selected based on the information and the receiving sub-codebook selected based on the designation of the selection are used to specify the first sub-codebook by the designation in the input signal. Be A first inverse vector quantizer for selecting a Coutl codeword, and a second
Second inverse vector quantizing means for selecting the vector code word of the first and second first vector selecting means selected by the first inverse vector quantizing means.
And the control means for controlling so as to select the decoded vector that gives a small quantization distortion by comparing the vector code word of 1) with the second vector code word selected by the second dequantization means. A quantized inverse quantization system including a quantizer. 11. A first vector code word and a second vector quantum corresponding to a first vector quantizing means adapted to a characteristic of an input signal in advance, instead of a transmission code book composed of a plurality of transmission sub-codebooks. Transmission codebook accommodating a plurality of second vector codewords corresponding to the conversion means, and a part of the codebook by changing the ratio of the number of words of the first vector codeword to the number of words of the second vector codeword And a transmission window moving means for cutting out the transmission sub codebook, the transmission sub codebook selecting means extracts information on the power of the input signal, and based on the extracted information, sets the transmission window moving means to a predetermined value. A transmission sub-codebook selecting means for moving to a position to obtain a predetermined transmission sub-codebook, and the first vector quantizing means and the second vector quantizing means perform quantization based on the predetermined sub-codebook. With the quantization means , A first vector code word corresponding to a first inverse vector quantizing means and a second inverse vector quantizing means corresponding to a characteristic of an output signal in advance, instead of a reception code book composed of a plurality of reception sub-codebooks A received codebook containing a plurality of second vector codewords, the number of words of the first vector codeword, and the second vector codeword
And a reception window moving means for cutting out a part of the reception codebook by changing the ratio of the number of vector codewords to form a reception sub-codebook, and the reception sub-codebook selecting means determines the power of the transmitted input signal. Related information is extracted, and the reception window moving means is moved to a predetermined position based on the extracted information to form a reception sub-codebook selecting means for obtaining a predetermined reception sub-codebook. 11. The quantized inverse quantization system according to claim 10, wherein the second inverse vector quantization means is an inverse quantization device adapted to perform inverse quantization based on the predetermined reception sub-codebook.