JP2002073097A - Celp type voice coding device and celp type voice decoding device as well as voice encoding method and voice decoding method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To perform the high-quality encoding of an incoming signal by improving the encoding performance of a fixed code note. SOLUTION: An LPC(Linear Prediction Coding) analyzer 101 performs the linear prediction analysis of an inputted voice signal, and an LPC quantizer 102 quantizes a linear predictive coefficient obtained by the analysis. A parameter selector 113 selects an sound source signal whose signal synthesized by a synthetic filter 208 is closest to the inputted voice signal from among an adaptive code storage area according to the quantized linear predictive coefficient, and then performs an operation to select an sound source signal that reduces the error between the synthetic signal and the inputted voice signal from a fixed code note 203 in this order. A feature extraction device 114 extracts the feature of an sound source signal to be generated from the fixed code note 203 by using the search result of the inputted voice signal, the quantization linear predictive coefficient and the adaptive code storage area, operates the fixed code note 203 so that it has the extracted feature of the sound source signal, and then searches the fixed code note 203.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a mobile station device such as a mobile phone or an information communication terminal device having a mobile phone function and a computer function in a mobile communication system, and a base station device performing wireless communication with the mobile station device. A CELP-type speech encoding apparatus and a CELP-type speech, which are provided with a low bit rate speech encoding function for encoding and transmitting a speech signal, and which expresses the speech signal separately into vocal tract information and sound source information. The present invention relates to a decoding device, an audio encoding method, and an audio decoding method.

[0002]

2. Description of the Related Art In the field of digital mobile communication and voice storage, a voice coding apparatus for compressing voice information for efficient use of radio waves and storage media and coding the voice information with high efficiency is used. .

[0003] Among them, CELP (Code Excited Linear Pre
diction (Code Excited Linear Predictive Coding) is widely used at medium and low bit rates.

[0004] Regarding the CELP technology, MRSchroede
r and BSAtal: "Code-Excited Linear Prediction
(CELP): High-quality Speech at Very Low Bit Ra
tes ", Proc. ICASSP-85, 25.1.1, pp. 937-940, 1985".

[0005] In the CELP type speech coding system, speech is divided into a certain frame length (about 5 ms to 50 ms),
Linear prediction of speech is performed for each frame, and a prediction residual (excitation signal) by the linear prediction for each frame is encoded using an adaptive code vector having a known waveform and a noise code vector.

[0006] The adaptive code vector is selected from an adaptive code book that stores driving excitation vectors generated in the past, and the noise code vector is selected from a fixed code book that stores a predetermined number of vector patterns.

As the noise code vector stored in the fixed codebook, a random noise sequence vector, a vector generated by arranging some pulses at different positions, and the like are used.

In the conventional CELP coding apparatus, LPC analysis and quantization, pitch search, noise codebook search and gain codebook search are performed using the input digital signal, and the quantized LPC code (L), The pitch period (P), the noise codebook index (S) and the gain codebook index (G) are transmitted to the decoder.

[0009]

However, in the conventional apparatus, it is necessary to encode an input signal having various characteristics using one kind of noise codebook (fixed codebook). For this reason, there is a problem that there is a limit in performing high-quality encoding.

[0010] The present invention has been made in view of such a point, and can improve the coding performance of a fixed codebook.
Accordingly, an object of the present invention is to provide a CELP-type speech encoding device, a CELP-type speech decoding device, a speech encoding method, and a speech decoding method that can encode a high-quality input signal.

[0011]

A CELP-type speech coding apparatus according to the present invention uses analysis means for performing linear prediction analysis of an input speech signal, and directly or indirectly uses the linear prediction coefficient obtained by this means. A filter constructed as described above, at least one or more codebooks for generating a source signal which is a signal for driving the filter and having a different shape / property / characteristic according to the characteristics of the source signal; A feature extraction unit for extracting a feature of the excitation signal using a part or all of the signal, the result of the linear prediction analysis, and at least one piece of information of the codebook.

According to this configuration, since a codebook of the excitation that positively uses the characteristics of the excitation signal is provided, excitation coding with higher quality can be realized.

[0013] In the CELP-type speech coding apparatus according to the present invention, in the above configuration, a quantization means for quantizing the linear prediction coefficient, an adaptive codebook for holding an excitation signal generated in the past,
A fixed codebook for storing a predetermined number of signal patterns and generating a sound source signal having a different shape / property / feature in accordance with the characteristics of the sound source signal; A configuration is employed in which the characteristics of the excitation signal are extracted using a part or all of the information of the linear prediction analysis result, the quantized linear prediction coefficient, and the adaptive codebook.

According to this configuration, the noise code vector output from the fixed codebook has a characteristic that a noise code vector extracted in advance has to have. A codebook can be realized, and a high-quality CELP-type speech coding apparatus can be realized.

A CELP-type speech decoding apparatus according to the present invention comprises: a linear prediction coefficient decoding means for decoding a linear prediction coefficient from an encoded speech signal; a filter constructed using the linear prediction coefficient obtained by the means; At least one or more codebooks for generating an excitation signal having a different shape / property / characteristic according to the characteristics of the excitation signal to the filter, and excitation signal decoding means for decoding the excitation signal from the at least one or more codebooks And the received encoded excitation signal characteristic information or
A feature decoding unit that decodes a feature of the excitation signal using the decoded linear prediction coefficient and decoding information of at least one of the codebooks.

According to this configuration, since a codebook of the excitation that positively utilizes the characteristics of the excitation signal is provided, it is possible to realize higher-quality excitation decoding.

In the CELP type speech decoding apparatus according to the present invention, the adaptive codebook for holding the previously decoded excitation signal, a predetermined number of signal patterns, and the characteristics of the excitation signal are stored. Different shape / property /
A fixed codebook that generates an excitation signal having a characteristic, and the characteristic decoding means uses the received encoded excitation signal characteristic information or the decoded linear prediction coefficient and the decoded adaptive codebook information. A configuration for decoding characteristics of the sound source signal is employed.

According to this configuration, the noise code vector output from the fixed codebook has a characteristic that the noise code vector extracted in advance should have, so that a high-quality fixed codebook is obtained. A codebook can be realized, and a high-quality CELP-type speech decoding device can be realized.

A speech encoding / decoding apparatus according to the present invention comprises a CELP type speech encoding apparatus and a CEL
An audio encoding / decoding device comprising a P-type audio decoding device.

According to this configuration, in the speech coding / decoding device, it is possible to obtain the same operation and effects as those of the CELP-type speech coding device and the CELP-type speech decoding device having the same configuration as any of the above.

The storage medium of the present invention performs a linear prediction analysis of the input speech signal, quantizes the linear prediction coefficient obtained by this analysis, and converts the synthesized signal into the input speech signal in accordance with the quantized linear prediction coefficient. In addition to selecting an excitation signal closest to the signal from the adaptive code storage area and sequentially selecting an excitation signal from the fixed code storage area to reduce the error between the synthesized signal and the input speech signal, the operation is sequentially performed. Extracting the characteristics of the excitation signal to be generated from the fixed code storage area using the input speech signal, the quantized linear prediction coefficient, and the search result of the adaptive code storage area, and extracting the characteristics of the extracted excitation signal. After performing an operation on the fixed code storage area so as to have the following configuration, software for voice encoding processing for searching the fixed code storage area is stored.

According to this configuration, in the storage medium, the same operation and effect as any of the above can be obtained.

The mobile station apparatus of the present invention employs a configuration including any one of the CELP type speech coding apparatus, CELP type speech decoding apparatus, speech coding / decoding apparatus, and storage medium having the same configuration as any of the above. .

According to this configuration, in the mobile station device,
The same operation and effect as any of the above can be obtained.

The base station apparatus of the present invention employs a configuration including any one of the CELP-type speech coding apparatus, CELP-type speech decoding apparatus, speech coding / decoding apparatus, and storage medium having the same configuration as any of the above. .

According to this configuration, in the base station apparatus,
The same operation and effect as any of the above can be obtained.

The speech coding method of the present invention performs a linear prediction analysis of an input speech signal, quantizes the linear prediction coefficients obtained by the analysis, and generates a synthesized signal according to the quantized linear prediction coefficients. The operation of selecting the sound source signal closest to the input speech signal from the adaptive code storage means and selecting the sound source signal for reducing the error between the synthesized signal and the input speech signal from the fixed code storage means is sequentially performed. Done to
Using the input speech signal, the quantized linear prediction coefficient, and the search result of the adaptive code storage unit, extract the characteristics of the excitation signal to be generated from the fixed code storage unit, and extract the characteristics of the extracted excitation signal. After performing an operation on the fixed code storage means as described above, the fixed code storage means is searched.

According to this method, since a codebook of the excitation that positively utilizes the characteristics of the excitation signal is provided, excitation coding of higher quality can be realized.

According to the speech decoding method of the present invention, a decoded speech signal is output by decoding a linear prediction coefficient from an encoded speech signal and filtering the excitation signal in accordance with the linear prediction coefficient obtained by the decoding. Different shapes / properties / characters from at least one or more code storage units according to the characteristics of the excitation signal.
Generating the excitation signal having characteristics, decoding the generated excitation signal, and receiving the encoded excitation signal characteristic information or the decoded linear prediction coefficient and decoding information of at least one of the code storage means; , The characteristic of the sound source signal is decoded.

According to this method, since a codebook of the excitation that actively utilizes the characteristics of the excitation signal is provided, it is possible to realize higher-quality excitation decoding.

[0031]

Embodiments of the present invention will be described below in detail with reference to the drawings.

(Embodiment 1) FIG. 1 is a block diagram showing a configuration of a CELP-type speech coding apparatus according to Embodiment 1 of the present invention.

A CELP type speech coding apparatus 100 shown in FIG. 1 comprises an LPC analyzer 101 and an LPC quantizer 10
2, a synthesis filter 103, and a parameter generator 104
, Adaptive codebook 105, fixed codebook 106, gain codebook 107, multiplier 108, adder 109, multiplier 110, adder 111, auditory weighting filter 1
12, a parameter selector 113, and a feature extractor 114
It is comprised including.

The operation of such a configuration will be described.

Input data consisting of digitized audio signals and the like is input to the LPC analyzer 101, adder 111 and feature extractor 114. As the input data, data obtained by cutting a DC component, limiting a band, and the like using a high-pass filter, a band-pass filter, or the like is generally used.

The LPC analyzer 101 performs a linear prediction analysis to calculate a linear prediction coefficient (LPC), and
Output to the PC quantizer 102.

The LPC quantizer 102 receives the input LP
C is quantized, and the LPC after quantization is
And the feature extractor 114, and the code L representing the quantized LPC is output to a CELP-type speech decoding apparatus described in a second embodiment described later. However, LPC quantization is generally performed by converting to LSP (Line Spectrum Pair) having good interpolation characteristics.

The synthesis filter 103 obtains a synthesized signal by performing a filtering process on the drive excitation signal output from the adder 109 in accordance with the quantized LPC, and outputs the synthesized signal to the adder 111.

The adder 111 calculates an error between the input data and the synthesized signal, and outputs the calculated error to the auditory weighting filter 112.

The auditory weighting filter 112 is provided by the adder 1
The error obtained in step 11 is perceptually weighted and output to the parameter selector 113.

Parameter selector 113 outputs fixed codebook index F output from parameter generator 104.
j, out of the combinations of the adaptive codebook index Ai and the gain codebook index Gk, the combination (A, F,
G) is selected and output to the CELP type speech decoding device.

The parameter generator 104 converts the fixed codebook index Fj, the adaptive codebook index Ai and the gain codebook index Gk into the fixed codebook 106, the adaptive codebook 105, the gain codebook 107 and the parameter selector 1.
13 to provide three codebooks 105-1
07 is controlled.

The adaptive codebook 105 buffers the driving excitation signal generated in the past while sequentially updating the driving excitation signal.
An adaptive code vector is generated using the adaptive codebook index {pitch period (pitch lag)} Ai input from parameter generator 104. The adaptive code vector generated by adaptive code book 105 is output to adder 109 after being multiplied by an adaptive code book gain by multiplier 108.

The feature extractor 114 uses all or a part of the input data, the result of the linear prediction analysis, the result of the quantization of the linear prediction coefficient, and the result of the pitch analysis (output of the adaptive codebook) to obtain the feature that the noise code vector should have. Is extracted. For example, by passing the input data through an inverse filter constructed in accordance with the quantized linear prediction coefficients to obtain an ideal driving excitation signal, and subtracting a signal obtained by multiplying the adaptive code vector by an appropriate gain from the ideal driving excitation signal. , An ideal signal is obtained as a noise code vector.

If the obtained ideal noise code vector is further subjected to linear prediction analysis or the like, spectrum envelope information or the like of the ideal noise code vector can be obtained, and this can be defined as a feature that the noise code vector should have. Conceivable. The feature information thus extracted is output to fixed codebook 106. If transmission to the CELP-type speech decoding device is indispensable, the feature information is transmitted to the CELP-type speech decoding device.

Fixed codebook 106 stores a predetermined number of different random code vectors having different shapes, and outputs a random code vector specified by index Fj of the random code vector input from parameter generator 104. I do.

Further, the fixed codebook 106 includes the feature extractor 1
It has a structure in which the shape / property / feature of the output noise code vector is changed using the feature information input from. For example, the feature information is classified (clustered) into several types, different fixed codebooks are assigned to the respective classes, and the fixed codebook is switched according to the classification, or the feature information is the spectral envelope information of the ideal noise code vector. In such a case, a process of applying a filter having the spectral envelope characteristic may be performed.

The gain codebook 107 stores a predetermined number of sets (gain vectors) of the adaptive codebook gain and the fixed codebook gain. The gain codebook 107 uses the gain codebook index Gk input from the parameter generator 104. Multiplier 108 multiplies the adaptive codebook gain component of the specified gain vector
Then, the fixed codebook gain components are output to multipliers 110, respectively.

If the gain codebook 107 has a multistage configuration, it is possible to reduce the amount of memory required for the gain codebook and the amount of calculation required for searching the gain codebook. Also, the gain codebook 10
If the number of bits allocated to 7 is sufficient, the adaptive codebook gain and the fixed codebook gain can be scalar-quantized independently.

The adder 109 includes multipliers 108 and 110
To generate a driving excitation signal by adding the adaptive code vector after each gain multiplication and the noise code vector input from
Output to synthesis filter 103 and adaptive codebook 105.

As described above, according to CELP-type speech coding apparatus 100 of the first embodiment, features to be included in the noise code vector output from fixed codebook 106 are extracted before feature codebook search. The fixed codebook 106 is adaptively constructed based on the feature information extracted by the extracting unit 114, so that high-quality noise code vector coding can be realized. In other words, the coding performance of the fixed codebook 106 can be improved, and thereby, high-quality input data (input speech) can be coded.

(Embodiment 2) FIG. 2 is a block diagram showing a configuration of a CELP-type speech decoding apparatus according to Embodiment 2 of the present invention.

CELP-type speech decoding apparatus 200 shown in FIG. 2 performs linear prediction coefficient coding information L output from CELP-type speech coding apparatus 100 described in the first embodiment and adaptive code vector coding. The information A, the random code vector coded information F, the gain coded information G, and the random code vector feature information C are input, and the LPC decoder 2
01, adaptive codebook 202, fixed codebook 203, gain codebook 204, multipliers 205 and 206, and adder 2
07 and a synthesis filter 208.

The operation of such a configuration will be described.

The LPC decoder 201 decodes a linear prediction coefficient using the input encoded information L, and
8 is output.

Adaptive codebook 202 decodes an adaptive codevector using input adaptive codevector coding information {generally, adaptive codebook lag (pitch) information} A, and outputs the result to multiplier 205.

The fixed codebook 203 stores the inputted random code vector coding information F and random code vector feature information C
Is used to decode the random code vector from the fixed codebook and output to the multiplier 206. When transmission of the noise code vector feature information C is not indispensable (when calculation can be performed in the decoder), the noise code vector feature information C is obtained by providing a feature extractor having the same configuration as the encoder side. Output to the multiplier 206.

The gain codebook 204 decodes the adaptive code vector gain and the noise code vector gain using the input gain coding information G, and multipliers 205 and 206 respectively.
Output to

Multiplier 205 multiplies the adaptive code vector input from adaptive codebook 202 by the adaptive code vector gain input from gain codebook 204, and outputs the multiplication result to adder 207.

Multiplier 206 multiplies the noise code vector input from fixed codebook 203 by the noise code vector gain input from gain codebook 204, and outputs the multiplication result to adder 207.

The adder 207 generates a driving excitation signal by adding the adaptive code vector after each gain multiplication and the noise code vector input from the multipliers 205 and 206, and generates the excitation signal and the synthesis filter 208 and the adaptive codebook 202. Output to

The synthesis filter 208 receives the input decoded L
According to the PC, a synthesized signal (decoded voice) is obtained by performing a filtering process on the drive excitation signal output from the adder 207, and this is output.

The synthesized signal is output as a decoded speech here. In general, a post-filter that has been subjected to post-processing such as formant emphasis and pitch emphasis to improve subjective quality by a post filter is subjected to final decoding. It is common to output as audio.

As described above, according to CELP-type speech decoding apparatus 200 of the second embodiment, since the excitation codebook that actively uses the characteristics of the excitation signal is provided, higher-quality excitation decoding is performed. Can be realized.

Further, since the noise code vector output from the fixed codebook has a configuration which has been extracted in advance and has the characteristics that the noise code vector should have, a high quality fixed codebook can be realized. High quality CE that can
An LP-type speech decoding device can be realized.

It is to be noted that a configuration may be adopted in which CELP-type speech encoding apparatus 100 of the first embodiment and CELP-type speech decoding apparatus 200 of the second embodiment are combined.

[0067]

As described above, according to the present invention,
It is possible to improve the coding performance of the fixed codebook, and thereby perform high-quality input signal coding.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a CELP-type speech coding apparatus according to Embodiment 1 of the present invention.

FIG. 2 is a block diagram showing a configuration of a CELP-type speech decoding device according to Embodiment 2 of the present invention.

[Explanation of symbols]

 REFERENCE SIGNS LIST 100 CELP-type speech coding apparatus 101 LPC analyzer 102 LPC quantizer 103, 208 synthesis filter 104 parameter generator 105 adaptive codebook 106 fixed codebook 107 gain codebook 108, 110 multiplier 109, 111 adder 112 auditory weighting Filter 113 parameter selector 114 feature extractor 200 CELP type speech decoder 201 LPC decoder 202 adaptive codebook 203 fixed codebook 204 gain codebook 205,206 multiplier 207 adder

Claims (10)

[Claims] An analysis means for performing a linear prediction analysis of an input speech signal, a filter constructed by directly or indirectly using a linear prediction coefficient obtained by the means, and a signal for driving the filter. And at least one or more codebooks for generating excitation signals having different shapes / properties / characteristics in accordance with the characteristics of the excitation signals, the input speech signal, the result of the linear prediction analysis, and at least one of the codebooks. Characteristic extracting means for extracting a characteristic of the sound source signal by using a part or all of the information.
Type audio coding device. 2. A quantization means for quantizing a linear prediction coefficient, an adaptive codebook for holding an excitation signal generated in the past,
A fixed codebook for storing a predetermined number of signal patterns and generating a sound source signal having a different shape / property / feature in accordance with the characteristics of the sound source signal; The CELP-type speech code according to claim 1, wherein the feature of the excitation signal is extracted using a part or all of the information of the quantized linear prediction coefficient and the adaptive codebook as a result of the linear prediction analysis. Device. 3. A linear prediction coefficient decoding means for decoding a linear prediction coefficient from an encoded speech signal, a filter constructed using the linear prediction coefficient obtained by the means, and a shape different according to the characteristics of the sound source signal. At least one or more codebooks for generating an excitation signal having a property / characteristic to the filter, excitation signal decoding means for decoding an excitation signal from the at least one or more codebooks, and a received encoded excitation signal Characteristic decoding means for decoding characteristics of the excitation signal using characteristic information or the decoded linear prediction coefficients and decoding information of at least one of the codebooks. Decryption device. 4. An adaptive codebook for holding an excitation signal decoded in the past, and storing a predetermined number of signal patterns, and generating an excitation signal having a different shape / property / characteristic according to the characteristic of the excitation signal. And a fixed codebook to generate,
The characteristic decoding means decodes the characteristic of the excitation signal using the received encoded excitation signal characteristic information or the decoded linear prediction coefficient and the decoded adaptive codebook information. The CELP-type speech decoding device according to claim 1. 5. The CELP-type speech coding apparatus according to claim 1 or claim 2, and the CE according to claim 3 or claim 4.
An audio encoding / decoding device comprising an LP type audio decoding device. 6. A linear prediction analysis of an input speech signal is performed, a linear prediction coefficient obtained by the analysis is quantized, and a synthesized signal is closest to the input speech signal according to the quantized linear prediction coefficient. The sound source signal is selected from the adaptive code storage area, and the operation of selecting the sound source signal for reducing the error between the synthesized signal and the input voice signal from the fixed code storage area is sequentially performed. Using the quantized linear prediction coefficient and the search result of the adaptive code storage area, extract the characteristics of the excitation signal to be generated from the fixed code storage area, and extract the characteristics of the extracted excitation signal to have the characteristics of the extracted excitation signal. A storage medium for storing voice encoding software for performing a search on the fixed code storage area after performing an operation on the fixed code storage area. 7. The CELP-type speech coding apparatus according to claim 1, wherein the CELP speech coding apparatus according to claim 3 or claim 4.
A mobile station device comprising: a voice decoding device, a voice coding / decoding device according to claim 5, and a storage medium according to claim 6. 8. The CELP-type speech coding apparatus according to claim 1 or 2, and the CELP according to claim 3 or 4.
A base station apparatus, comprising: a voice decoding apparatus, a speech coding / decoding apparatus according to claim 5, and a storage medium according to claim 6. 9. A linear prediction analysis of an input speech signal is performed, a linear prediction coefficient obtained by the analysis is quantized, and a synthesized signal is closest to the input speech signal according to the quantized linear prediction coefficient. The sound source signal is selected from the adaptive code storage means, and the operation of selecting the sound source signal for reducing the error between the synthesized signal and the input voice signal from the fixed code storage means is sequentially performed. Using the quantized linear prediction coefficients and the search results of the adaptive code storage means, extract characteristics of the excitation signal to be generated from the fixed code storage means, and extract the characteristics of the extracted excitation signal so as to have the characteristics of the extracted excitation signal. A speech encoding method comprising: performing a search on the fixed code storage unit after performing an operation on the fixed code storage unit. 10. A decoded speech signal is output by decoding a linear prediction coefficient from an encoded speech signal, filtering an excitation signal according to the linear prediction coefficient obtained by the decoding, and outputting at least one or more codes. The excitation signal having a different shape / property / characteristic is generated from the storage means according to the characteristic of the excitation signal, the generated excitation signal is decoded, and the received encoded excitation signal characteristic information or the decoded encoded excitation signal characteristic information is decoded. A speech decoding method characterized by decoding characteristics of the excitation signal using a linear prediction coefficient and decoding information of at least one code storage unit.