JP2796408B2 - Audio information compression device - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an audio information compression apparatus that compresses an audio signal for transmission or storage.

[Prior Art] Generally, an audio signal has a considerably high correlation between adjacent samples.
A means for efficiently compressing information of a speech signal by using a spectrum prediction method based on a signal sequence of neighboring samples is widely used. The voice signal also has repetitive redundancy depending on the pitch, and means for further improving the efficiency of information compression by pitch prediction is widely used.

As a low bit rate speech coding method using these two linear prediction means, CELP (Code-Excited Linear Prediction) is used.
radiction)) is being actively researched and developed (for example,
"Stochastic Coding of Speech
Signal at Very Low Bit Rate's The Importance of Speech Perception (Stochastic Coding of Speech Signals at Very Low)
Bit Rates: The Importance of Speech Perceptio
n) ", M. Schroeder and BS
Atal (MRShoreder and BSAtal), Speech Communication, 4, 1985, North-Holland, pp. 155-162).

In the CELP, the prediction residual signal is treated as a vector having a length of 40 samples, and the prediction residual signal is encoded by compressing the prediction residual signal at a very low bit rate using a vector quantization technique.

FIG. 5 shows a CELP as an example of a conventional audio information compression apparatus.
6 is shown, and FIG. 6 shows a decoder corresponding to the encoder shown in FIG.

 Hereinafter, the encoder of FIG. 5 will be described.

In the encoder shown in FIG. 5, an AD converter 51 converts an analog audio signal waveform input from an input terminal into digital data, and converts the waveform into a vector S having a fixed sample length (for example, 5 msec). The signals are output to the subtracters 58, respectively.

The analyzer 52 performs a linear prediction analysis on the vector S output from the AD converter 51. The prediction coefficient vector α corresponding to the vector S input as the analysis result and the average amplitude g of the residual signal are calculated, the prediction coefficient vector α is input to the linear predictor 53, and the average amplitude g of the residual signal is output. Multiplier 54
And the prediction coefficient vector α and the average amplitude g of the residual signal are also output to the multiplexer 55.
Here, the linear prediction is a generic term for spectrum prediction based on a signal sequence of neighboring samples and pitch prediction generally called long-term prediction.

The codebook 56 has a residual signal waveform of a fixed sample length and a normalized gain (hereinafter referred to as a codeword).
Are stored in a plurality of types (for example, 1024 types), and the stored code work is output to the multiplier 54 in accordance with the instruction of the optimum codeword selector 57.

The multiplier 54 converts the codeword output from the codebook 56 into an average amplitude g of the residual signal output from the analyzer 52.
, And outputs the result to a synthesis filter including the linear predictor 53 and the adder 61.

The synthesis filter receives the amplified codeword and outputs a synthesized signal vector S ′.

The subtractor 58 receives the composite signal vector S ′ and the input vector S and calculates a difference.

The auditory weighting filter 59 outputs the difference output from the subtractor 58 to the power calculator 60 in consideration of the effect of auditory spectral masking, the power is calculated by the power calculator 60, and the optimal codeword is selected. Sent to the vessel 57.
The optimum codeword selector 57 selects the codeword having the minimum power in the codebook 56 and outputs an index value index for specifying the codeword to the multiplexer 55.

The multiplexer 55 transmits the prediction coefficient vector α, the average amplitude g of the residual signal, and the index value index for specifying the residual signal.

Here, as a method of creating the codebook 56 for storing codewords, a method of analyzing a real audio signal for a long time to obtain a residual signal and creating the residual signal by vector quantization, or a method of white-Gaussian (White-Gaussia
There is a method of making a pseudo signal with a random signal of n).

 Next, the decoder shown in FIG. 6 will be described.

The demultiplexer 62 decomposes the signal received from the transmission line into a prediction coefficient vector α, an average amplitude g of the residual signal, and an index value index for specifying the residual signal.

The codebook 63 has the same contents as the codebook 56 of the encoder, and outputs a codeword to the multiplier 64 based on the index value index for specifying the residual signal.

The multiplier 64 amplifies the codeword received from the codebook 63 by the average amplitude g output from the demultiplexer 62.

The synthesis filter including the adder 65, the linear predictor 66, and the like calculates a synthesized signal vector using a synthesis filter formed according to the received prediction vector α.

The DA converter 67 receives the synthesized signal vector output from the synthesis filter, converts the reproduced signal into an analog signal,
Plays the input audio signal.

[Problems to be Solved by the Invention] However, in the above-described speech encoding method using CELP, the content of the codebook is one type, and the content is set so as to be optimized based on the result of long-term statistical analysis of speech. Have been. Therefore, it is known that the codebook of the results of long-term statistical analysis is not optimal, for example, at the time of the burst of a burst consonant, at the transition from silence and unvoiced to stationary vowels, etc. In general CELP expressed by, the residual signal in the nasal interval contains information on the zero point of the spectrum that could not be expressed by the all-pole model, and the result of the long-term statistical analysis There is a problem that codebooks cannot cope sufficiently.

An object of the present invention is to provide a plurality of residual signal codebooks suitable for various short-time voice patterns in view of the above-mentioned problems of the conventional voice information compression apparatus, and to provide an index obtained by vector-quantizing an analyzed prediction coefficient. It is an object of the present invention to provide an audio information compression apparatus which can select a codebook to be used from among a plurality of residual signal codebooks based on a value and efficiently compress audio information.

[Means for Solving the Problems] The above-described object of the present invention is to provide an analysis unit that performs a linear prediction analysis of an input speech signal and outputs an analysis result, and a first code that stores a plurality of linear prediction coefficient vectors. Storage means, vector quantization means for obtaining an optimal codeword index value from the first code storage means, a plurality of second code storage means each storing a plurality of residual signal vectors,
Selecting means for selecting a code storing means to be used from a plurality of second code storing means based on an index value of an optimum code word outputted from the vector quantizing means; and selecting all code storing means from the code storing means selected by the selecting means. A synthesizing filter unit that inputs a codeword and outputs a synthesized signal; and a multiplexer that outputs an index value for specifying an optimal codeword index value, an average amplitude of a residual signal, and a residual signal, and The two-code storage means stores, for each codeword of the linear prediction coefficient vector output from the vector quantization means, a code of the initial residual signal from an initial residual signal codebook comprising a large number of codewords suitable for various voice patterns. The contribution calculated based on the frequency of use of the word is calculated, and the code of the linear prediction coefficient vector output from the vector quantization means is calculated. For words, grouping similar pattern of contribution is achieved by an audio information compression apparatus characterized by stored by selecting a high contribution codeword from among them.

[Action]

The analysis means performs a linear prediction analysis of the input speech signal and outputs an analysis result, and the vector quantization means obtains an index value of an optimal codeword from the first code storage means storing a plurality of linear prediction coefficient vectors. , And a code storage used by the selection means from a plurality of second code storage means each storing a plurality of residual signal vectors based on an index value of an optimum code word output from the vector quantization means. Means, the synthesis filter means inputs all the codewords from the code storage means selected by the selection means and outputs a synthesized signal, and the multiplexer provides the index value of the optimum codeword, the average amplitude of the residual signal and An index value for specifying the residual signal is output. A plurality of second code storage means stores, for each codeword of the linear prediction coefficient vector output from the vector quantization means, an initial residual signal from an initial residual signal codebook comprising a large number of codewords suitable for various voice patterns. Calculate the contribution calculated based on the frequency of use of signal codewords, and summarize similar contribution patterns for codewords of the linear prediction coefficient vector output from the vector quantization means. Since a high-degree code word is selected and stored, it is possible to save the memory and improve the synthesis quality and to efficiently compress the audio signal.

[Embodiment] Hereinafter, an embodiment of a voice information compression device according to the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing a configuration of an encoder as a first embodiment of the audio information compression apparatus of the present invention, and FIG. 2 is a decoder corresponding to the encoder shown in FIG. FIG. 3 is a block diagram showing the configuration of FIG.

An encoder shown in FIG. 1 includes an AD converter 11, an analyzer 12 connected to the AD converter 11 as analysis means, and a predictive coefficient vector quantizer connected to the analyzer 12 as vector quantization means. In the figure, a prediction coefficient VQ 13, a codebook selector 14 as selection means connected to the prediction coefficient vector quantizer 13, a prediction coefficient vector quantizer 13 and a codebook selector 14 A linear predictor 15, a predictive coefficient vector quantizer 13, and a predictive coefficient codebook 16 as a first code storage connected to the linear predictor 15; a residual signal codebook 17 as a second code storage; 12
And a multiplier 18 connected to the residual signal codebook 17, an adder 19 connected to the multiplier 18 and the linear predictor 15, an AD converter 11, a subtractor connected to the adder 19 and the linear predictor 15. Two
0, an auditory weighting filter 21, connected to a subtractor 20,
Power calculator 2 connected to auditory weighting filter 21
2. It comprises an optimal codeword selector 23 connected to the power calculator 22, and a multiplexer 24 connected to the analyzer 12, the predictive coefficient vector quantizer 13 and the optimal codeword selector 23.

 Next, the operation of the above encoder will be described in detail.

First, the AD converter 11 converts the analog audio signal waveform input from the input terminal into digital data, and converts it into a vector S having a fixed sample length (for example, 5 msec).
And to the subtractor 20.

The analyzer 12 performs linear prediction analysis on the vector S output from the AD converter 11. Here, the linear prediction is a generic term for spectrum prediction based on a signal sequence of neighboring samples and pitch prediction generally called long-term prediction.
Then, the analyzer 12 calculates the prediction coefficient vector α and the average amplitude g of the residual signal as the analysis result, and outputs the prediction coefficient vector α to the prediction coefficient vector quantizer 13 to calculate the average amplitude g of the residual signal. Output to the multipliers 18 respectively.

The prediction coefficient vector quantizer 13 refers to the prediction coefficient codebook 16 and obtains an index value j of an optimum codeword.

The codebook selector 14 is a predictive coefficient vector quantizer.
A codebook to be used is selected from a plurality of residual signal codebooks 17 based on the index value j output from 13. A method for generating the plurality of residual signal codebooks 17 and a method for selecting a codebook to be used will be described later.

The residual signal codebook 17 selected by the codebook selector 14 stores a plurality of types (for example, 256 types) of codewords in which the gain of a fixed sample length is normalized.

The optimal codeword selector 23 is the codebook selector 14
Is instructed to output to the multiplier 18 all the codewords stored in the selected residual signal codebook.

After amplifying the codeword output from the residual signal codebook selected by the multiplier 18, the linear predictor 15
A synthesis filter as a synthesis filter means configured as described above calculates a synthesized signal vector S ′ from the amplified codeword and outputs it to the subtractor 20.

The subtracter 20 obtains an error by subtracting the synthesized signal vector S ′ from the input vector S, and outputs the error to the auditory weighting filter 21. The auditory weighting filter 21 outputs the input error to the power calculator 22 in consideration of the effect of spectrum masking, the power is calculated by the power calculator 22, and the optimal codeword selector 23 is selected. An index value i for specifying a code word having the minimum power in the code book is obtained.

The multiplexer 24 transmits the index value j for specifying the prediction coefficient vector, the average amplitude g of the residual signal, and the index value i for specifying the residual signal to the communication path 25.

Next, creation of a plurality of codebooks 17 for storing residual signal vectors suitable for the features of the above-described audio signal will be described.

In the encoder of the present embodiment, the creation of a codebook is a very important issue, and if it is not possible to properly create a plurality of codebooks having a plurality of codewords, the encoding performance will be improved. Can not expect. There is also a disadvantage that the amount of memory for storing a plurality of codebooks increases.

Hereinafter, as an example of a method of creating a plurality of codebooks 17 having a plurality of codewords and a method of selecting a codebook to be used, a procedure in a case where an actual audio signal is analyzed and created for a long time will be described.

Step 1: In the same manner as the conventional procedure, a large-sized (for example, N = 4096 codewords) residual signal codebook is created by analyzing long-term speech. The index value of this codeword is represented by n (n = 1, 2,..., N) (where N is a positive integer).

Step 2: The long time speech is encoded again by the prediction coefficient vector quantizer 13 using the residual signal codebook created in step 1.

At this time, let the index value of the vector quantization of the linear prediction coefficient be j (j = 1, 2,..., M, and M be any positive integer representing the codebook size of the linear prediction coefficient), Contributions are obtained for all codewords of the residual signal for each index value j. Here, the contribution indicates how often a certain codeword is used, and how the performance (quality of synthesized sound) is at that time.

As an example, if the contribution of the codeword of the index value n of the residual signal in the index value j of the vector quantization of the linear prediction coefficient is C _n (j), C _n (j) = (index value of the residual signal) n
) × (average performance of synthesized sound using index value n of residual signal).

Here, the average performance is the ratio of noise to signal (signal
Toe noise ratio (Signal-to-Noise Ratio)
Is the average value.

Step 3: Having a codebook of residual signals for each index value j of vector quantization of all linear prediction coefficients is impractical in that a huge amount of memory is required. Therefore, for each index value j of the vector quantization of the linear prediction coefficient,
Collect codebooks with similar usage,
It is necessary to reduce the pattern of the contribution of the codeword of the residual signal to a small number.

Such a summary is based on the assumption that the contribution C _n (j) is regarded as a vector of N samples with a variable j, and K is set to a number smaller than M using a vector quantization method, where K types (here, K can be realized by obtaining data C _n (k) (k = 1, 2,..., K) of the degree of contribution of the codebook type of the residual signal. That is, _assuming that the contribution is C _n (k), k = 1, 2,.
<M.

At the time of this vector quantization, the index value j of the vector quantization of the linear prediction coefficient and k summed up by the vector quantization
Is stored. An example of this correspondence is shown in Table 1 (here, M is 1024 and K is 8).

Through the above-described steps 1 to 3, the usage status of the codebook of the residual signal is clustered for each feature of the audio signal.

Step 4: As the last step, the code book of the K kinds of residual signals is reduced in size.

Until step 3, the content of the codebook of the residual signal of each cluster is the same, but the contribution of the codeword differs for each cluster.

Therefore, the size of each codebook is reduced by extracting a codeword having a high contribution for each codebook. Here, the codebook size of the residual signal in step 1 was 4096 (index value is represented by 12 bits), but from the one with the highest contribution, it is 1024 (index value is 10bi).
(represented by ts).

According to the above-described procedure, an appropriate codebook of the residual signal is created for each feature of the audio signal, and the created codebook can be selected.

Next, a decoder corresponding to the encoder of FIG. 1 is shown in FIG.

The decoder of FIG. 2 is connected to a demultiplexer 26, a codebook selector 27 connected to the demultiplexer 26, a residual signal codebook 28 connected to the demultiplexer 26, a demultiplexer 26 and a codebook 28. Multiplier 29, an adder 30 connected to the multiplier 29, a linear predictor 31 connected to the adder 30, a prediction coefficient codebook 32 connected to the linear predictor 31, an adder 30, and a linear predictor 31 And a D / A converter 33 connected thereto.

 Next, the operation of the decoder shown in FIG. 2 will be described.

The demultiplexer 26 decomposes the signal received from the transmission path 25 into an index value j for specifying the prediction coefficient vector α, an average amplitude g of the residual signal, and an index value i for specifying the residual signal.

The codebook selector 27 selects a residual signal codebook to be used based on an index value j specifying a prediction coefficient vector α from a plurality of residual signal codebooks 28 provided in advance. This operation is the same as that of the encoder.
The plurality of residual signal codebooks 28 and the prediction coefficient codebook 32 have the same contents as the codebooks 17 and 16 of the encoder.

When the residual signal codebook to be used is selected by the codebook selector 27, the multiplier 29 multiplies the codeword specified by the index value i of the selected residual signal codebook.
Output to

The multiplier 29 receives the specified codeword, amplifies the codeword by the average amplitude g output from the demultiplexer 26, and outputs the amplified codeword to a synthesis filter including an adder 30, a linear predictor 31, and the like. I do.

The synthesis filter calculates a synthesized signal vector from the codeword amplified by the synthesis filter formed according to the index value j of the prediction coefficient codeword output from the demultiplexer 26, and outputs the resultant to the DA converter 33.

The DA converter 33 converts the input composite signal vector into an analog signal and reproduces the input audio signal.

By changing the codebook of the residual signal based on the characteristics of the audio signal by the above operation procedure, audio information can be efficiently compressed.

In the first embodiment, by providing a plurality of residual signal codebooks, the required memory amount increases. As can be understood from the procedure described in the method of generating a plurality of residual signal codebooks described above, each residual signal codebook is extracted from a large (eg, 4096) residual signal codebook (eg, 1024 types). ). Note that the contents of a plurality of residual signal codebooks include duplicate codewords.

Next, FIG. 3 shows a block diagram of a configuration of an encoder as a second embodiment of the audio information compression apparatus of the present invention, and FIG. 4 shows a decoder corresponding to the encoder shown in FIG. 1 shows a block diagram of the configuration of FIG.

In view of the above points in the first embodiment, the encoder shown in FIG. 3 is configured with a reduced amount of memory required for a plurality of residual signal codebooks.

The difference between the encoder shown in FIG. 3 and the decoder shown in FIG. 4 from the encoder shown in FIG. 1 and the decoder shown in FIG. Residual signal codebook
Instead of 17 and 28, a plurality of residual signal codebooks 34 and 36 having different contents are provided, and new residual signal codebooks 35 and 37 are added.

Except for the above-mentioned parts, the operation is exactly the same as that of the first embodiment. Therefore, only the residual signal codebooks 34, 35, 36, and 37 will be described below.

In the encoder shown in FIG. 3 and the decoder shown in FIG. 4, the only codebooks that actually store the residual signals are the residual signal codebooks 35 and 37. The plurality of residual signal codebooks 34 and 36 do not directly store the residual signals, but only store index values that specify the codewords of the residual signal codebooks 35 and 37. Therefore,
The memory required for the plurality of residual signal codebooks 34 and 36 is relatively small.

It should be noted that codebooks 35 and 3 for actually storing the residual signals are provided.
7 is obtained by creating the residual signal codebooks 34 and 36 in the above-mentioned first embodiment.
Are all listed up, and a codeword may be extracted and created from the codebook of the large residual signal created in step 1 so as not to overlap.

The encoder of FIG. 3 and the decoder of FIG. 4 efficiently change the codebook of the residual signal stored in a small amount of memory based on the characteristics of the audio signal, thereby efficiently generating audio information. Can be compressed.

[Effects of the Invention] Analysis means for performing a linear prediction analysis on an input speech signal and outputting an analysis result, first code storage means for storing a plurality of linear prediction coefficient vectors, and the first code storage means Vector quantization means for obtaining an index value of a proper codeword, a plurality of second code storage means each storing a plurality of residual signal vectors, and an index of the optimum codeword output from the vector quantization means. Selecting means for selecting a code storage means to be used from the plurality of second code storage means based on a value; and synthesizing for inputting all codewords from the code storage means selected by the selecting means and outputting a synthesized signal. Filter means, and a multiplexer for outputting an index value of the optimal code word, an average amplitude of the residual signal, and an index value for specifying the residual signal. , The plurality of second code storage means, for each codeword of a linear prediction coefficient vector output from the vector quantization means, from an initial residual signal codebook comprising a large number of codewords suitable for various audio patterns, Calculate the contribution calculated based on the frequency of use of the codeword of the initial residual signal, for the codeword of the linear prediction coefficient vector output from the vector quantization means,
Patterns with similar contributions are put together, and the codewords with the highest contributions are selected and stored from among them. As a result, it is possible to save the memory and improve the synthesis quality, and efficiently use the speech signal. Can be compressed.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of an encoder as a first embodiment of the audio information compression apparatus of the present invention, and FIG. 2 is a configuration example of a decoder corresponding to the encoder shown in FIG. FIG. 3 is a block diagram showing a configuration of an encoder as a second embodiment of the audio information compression apparatus of the present invention, and FIG. 4 is an encoding corresponding to the encoder shown in FIG. FIG. 5 is a block diagram showing an example of the configuration of a decoder, FIG. 5 is a block diagram showing an example of a decoder of a conventional audio information compression apparatus, and FIG. 6 is a block diagram of a decoder corresponding to the encoder shown in FIG. It is a block diagram showing an example. 11… AD converter, 12… Analyzer, 13… Predictor coefficient vector quantizer, 14,27 …… Codebook selector, 15,31 ……
Linear predictor, 16, 32 …… Predictive coefficient codebook, 17, 28, 3
4,36 …… Residual signal codebook, 18,29 …… Multiplier, 19,
30 adder, 20 subtracter, 21 auditory weighting filter, 22 power calculator, 23 optimal codeword selector, 24 multiplexer, 26 demultiplexer, 33 DA converter.

────────────────────────────────────────────────── ─── Continued on the front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G10L 3/00-9/18 H03M 7/30 JICST (JOIS)

Claims (1)

(57) [Claims] An analysis means for performing a linear prediction analysis of an input speech signal and outputting an analysis result; a first code storage means for storing a plurality of linear prediction coefficient vectors; Vector quantization means for obtaining an index value of a proper codeword, a plurality of second code storage means each storing a plurality of residual signal vectors, and an index of the optimum codeword output from the vector quantization means. Selecting means for selecting a code storage means to be used from the plurality of second code storage means based on a value; and synthesizing for inputting all codewords from the code storage means selected by the selecting means and outputting a synthesized signal. Filter means, and an index value of the optimal codeword, a multiplexer that outputs an average value of the residual signal and an index value that specifies the residual signal, The plurality of second code storage units are configured to store, for each codeword of the linear prediction coefficient vector output from the vector quantization unit, an initial residual signal codebook including a large number of codewords suitable for various audio patterns, The contribution calculated based on the frequency of use of the code words of the initial residual signal is calculated, and for the code words of the linear prediction coefficient vector output from the vector quantization means, patterns having similar contributions are summarized, An audio information compression apparatus characterized by selecting and storing a code word having a high contribution degree from among them.