JP2004198522A - Method of updating adaptive code book, voice encoding device, and voice decoding device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve the tone quality by updating an adaptive code book in a voice encoding/decoding device of the CELP system. <P>SOLUTION: A decoded waveform generation part 201 uses two sound sources corresponding to a gain code and an index to generate a local sound source waveform. A decoded waveform insertion part 202 shifts decoded sound source data stored in a decoded sound source data memory 204 from the future to the past to abandon a portion overflowing from the memory and stores a local decoded sound source waveform in a future empty part. A filter processing part 203 filters decoded sound source data having had the local decoded sound source waveform inserted thereto. Decoded sound source data after pitch filtering in the filter processing part 203 is taken as an updated adaptive code book. <P>COPYRIGHT: (C)2004,JPO&NCIPI

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a method for updating an adaptive codebook in a speech encoding device / decoding device of the CELP system.
[0002]
[Prior art]
When transmitting a voice signal in a packet communication system represented by the Internet communication, a mobile communication system, or the like, a compression encoding technique is often used in order to increase the transmission efficiency of the voice signal. Many speech coding schemes have been developed so far, and many of the recently developed low bit rate speech coding schemes such as the CELP scheme separate a speech signal into spectrum envelope information and spectrum fine structure information. This is a method of compressing and encoding the separated information (for example, Non-Patent Document 1).
[0003]
In recent years, there has been a demand for further improvement in sound quality of the speech encoding device. In this regard, research is progressing on the configuration of a probabilistic codebook, the configuration of an adaptive codebook, and a search method in a CELP-type speech coding apparatus (for example, Patent Document 1).
[0004]
[Patent Document 1]
Japanese Patent Application Laid-Open No. 2002-182697 [Non-Patent Document 1]
MRSchroeder, BSAtal, "Code Excited Linear Prediction: High QualitySpeech at Low Bit Rate", IEEE proc., ICASSP'85 pp.937-940
[0005]
[Problems to be solved by the invention]
However, little research has been done on updating the adaptive codebook, and there is a problem in improving the performance of this part.
[0006]
The present invention has been made in view of the above points, and an adaptive codebook updating method, a speech encoding apparatus, and a speech decoding apparatus capable of improving sound quality in a speech encoding apparatus / speech decoding apparatus of a CELP system. It is intended to provide a device.
[0007]
[Means for Solving the Problems]
The adaptive codebook updating method according to the present invention includes a step of memory-shifting decoded excitation data, which is an adaptive codebook, from the future to the past and discarding a portion coming out of the memory; A method including a step of storing the excitation waveform and a step of filtering the decoded excitation data after the local decoded excitation waveform is inserted is employed.
[0008]
With this method, the pitch structure of the adaptive codebook can be emphasized, and the decoded excitation data can be processed so as to obtain better sound quality.
[0009]
The method of updating the adaptive codebook of the present invention employs a method of using backward filtering in which filtering is performed from the future to the past in the step of performing filtering.
[0010]
By this method, the decoded excitation data can be processed so as to obtain better sound quality, and furthermore, transmission of transmission errors due to the processing can be reduced.
[0011]
In the adaptive codebook updating method according to the present invention, in the step of performing filtering, pitch filtering is performed by using a lag corresponding to the code of the adaptive codebook and a gain of the excitation of the decoded adaptive codebook by the pitch filter of the backward filter. Take the method of doing.
[0012]
By this method, decoded excitation data having a clear pitch structure can be prepared.
[0013]
The speech coding apparatus of the present invention employs a configuration for updating the adaptive codebook by any of the above methods. Further, the speech decoding apparatus of the present invention employs a configuration for updating an adaptive codebook corresponding to any of the above methods.
[0014]
With these configurations, the decoded excitation data, which is an adaptive codebook, can be processed so as to obtain better sound quality.
[0015]
BEST MODE FOR CARRYING OUT THE INVENTION
The gist of the present invention is that the decoded excitation data, which is an adaptive codebook, is shifted from the future to the past in memory and the portion that comes out of the memory is discarded, the local decoded excitation waveform is stored in a vacant portion in the future, That is, the adaptive codebook is updated by filtering the decoded excitation data after inserting the decoded excitation waveform.
[0016]
Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.
[0017]
(Embodiment 1)
FIG. 1 is a block diagram showing a configuration of a CELP-type speech encoding device. It is assumed that the input audio signal is sequentially input to the audio encoding device for each processing frame divided at a time interval of about 20 ms.
[0018]
The input audio signal input to the audio encoding device for each processing frame is first supplied to the LPC analysis unit 101. The LPC analysis unit 101 obtains LPC coefficients by analyzing an input speech signal by LPC (Linear Predictive Coding), quantizes the LPC coefficients into an LPC code, and decodes the LPC code to obtain decoded LPC coefficients.
[0019]
The adaptive codebook 102 generally stores past synthesized excitation waveforms in a memory for a fixed length (pitch search range). Here, the coding includes two values, a value of how far the synthetic sound source is used (generally called a lag, which corresponds to a fundamental period (pitch) for voiced sounds such as human vowels) and a gain thereof. This is done with parameters. Therefore, the adaptive codebook 102 greatly contributes to the coding efficiency in coding a portion where regularly similar waveforms such as human vowels are arranged.
[0020]
The probabilistic codebook 103 stores a plurality of excitation samples having a numbered subframe length.
[0021]
Excitation generating section 104 reads an adaptive code vector and a fixed code vector from adaptive codebook 102 and stochastic codebook 103, respectively, and sends them to LPC synthesis section 105. The LPC synthesizing unit 105 synthesizes and filters the adaptive code vector and the fixed code vector supplied from the sound source creating unit 104 with an all-pole synthesis filter having the decoded LPC coefficient provided from the LPC analyzing unit 101 as a filter coefficient. , A combined adaptive code vector and a combined fixed code vector.
[0022]
The comparing unit 106 analyzes the relationship between the synthesized adaptive code vector, the synthesized fixed code vector, and the input speech signal output from the LPC synthesizing unit 105, and calculates the adaptive codebook optimum gain by which the synthesized adaptive code vector is multiplied and the synthesized fixed code vector. To obtain the optimal gain of the probabilistic codebook multiplied by.
[0023]
Further, comparing section 106 adds a vector obtained by multiplying the synthesized adaptive code vector by the adaptive codebook optimum gain and a vector obtained by multiplying the synthesized fixed code vector by the probabilistic codebook optimum gain to obtain a synthesized speech vector. And calculates the distance between the synthesized speech and the input speech signal. Then, the comparison unit 106 obtains a synthesized speech vector for all combinations of the adaptive code vector stored in the adaptive codebook 102 and the fixed code vector stored in the stochastic codebook 103, and The index of the adaptive code vector and the index of the fixed code vector that minimize the distance to the signal are obtained. Then, the comparing unit 106 converts the index of the code vector output from each codebook, each code vector corresponding to the index, the adaptive codebook optimal gain and the stochastic codebook optimal gain corresponding to the index into the parameter encoding unit. Send to 107.
[0024]
The parameter encoding unit 107 acquires the gain code by encoding the adaptive codebook optimum gain and the stochastic codebook optimum gain, and obtains the gain code, the LPC code given from the LPC analysis unit 101, and the index of each codebook. Are output collectively for each processing frame.
[0025]
Further, parameter encoding section 107 includes a vector obtained by multiplying an adaptive codebook corresponding to the index of the adaptive codebook by an adaptive codebook gain corresponding to the gain code, and a fixed code vector corresponding to the index of the stochastic codebook. And a vector obtained by multiplying by the probabilistic codebook gain corresponding to the gain code to the adaptive codebook 102.
[0026]
Note that the synthesis filtering in the LPC synthesis unit 105 generally uses a linear prediction coefficient, a high-frequency emphasis filter, and an audibility weighting filter using a long-term prediction coefficient obtained by performing a long-term prediction analysis on the input speech. .
[0027]
In addition, the search for the optimal index of the adaptive codebook and the stochastic codebook, the calculation of the optimal gain, and the encoding processing of the optimal gain are generally performed in subframe units obtained by further dividing the frame.
[0028]
Next, the internal processing of adaptive codebook 102 will be described using FIG. FIG. 2 is a functional block diagram showing an internal configuration of adaptive codebook 102 according to the present embodiment. As shown in FIG. 2, adaptive codebook 102 mainly includes a decoded waveform generation unit 201, a decoded waveform insertion unit 202, a filter processing unit 203, and a decoded excitation data memory 204.
[0029]
The decoded waveform generation unit 201 generates a local excitation waveform D (i) using the following two excitations corresponding to the gain code and the index input from the parameter encoding unit 107 by the following equation (1).
D (i) = ga × A (i) + gs × S (i) (1)
i: Waveform index (i = 0 to L-1)
L: subframe length ga: excitation gain of decoded adaptive codebook A (i): excitation waveform of decoded adaptive codebook gs: excitation gain S (i of decoded probabilistic codebook ): The decoded excitation waveform of the probabilistic codebook and the generated local excitation waveform D (i) are output to the decoded waveform insertion unit 202.
[0030]
The decoded waveform inserting unit 202 shifts the decoded sound source data stored in the decoded sound source data memory 204 from the future to the past and discards the portion that came out of the memory. The converted sound source waveform D (i) is stored. The following equation (2) expresses this procedure.
M (i) = M (i + L) (where i = 0 to PL-1)
M (P−L + I) = D (i) (where i = 0 to L−1) (2)
M (i): Decoded excitation data as adaptive codebook
i: Waveform index (i = 0 to L-1)
L: Subframe length P: Memory length for storing decoded excitation data as an adaptive codebook
A signal indicating completion of the above processing is output to filter processing section 203. Filter processing section 203 performs filtering on the decoded excitation data after the local decoded excitation waveform D (i) has been inserted.
[0032]
Here, in the present embodiment, a case where backward filtering is used will be described. It should be noted that the effects of pitch enhancement, frequency adaptation, and the like can be obtained by forward filtering, but the propagation of errors is slightly longer.
[0033]
A pitch filter will be described as an example of a backward filter. The effect can be obtained by using a low-pass / high-pass emphasis filter, a band-pass filter, or the like other than the pitch filter.
[0034]
Filter processing section 203 performs pitch filtering using the lag corresponding to the code in the adaptive codebook and the gain of the excitation of the decoded adaptive codebook. The specific processing is shown in the following equation (3).
M (i) = M (i) + a × M (i + T) (where i = PT-1 to 0) (3)
M (i): Decoded excitation data as adaptive codebook a: Filter coefficient (a = 0.05 × ga + 0.09)
i: waveform index T: decoded lag ga: decoded adaptive codebook excitation gain
Note that, in the above, the constants “0.05” and “0.09” are linear constants representing the relationship between the decoded gain and the coefficient of the pitch filter. Values shown). By adjusting this constant, the strength of the pitch component of the adaptive codebook can be changed. Of course, it is easy to improve the performance by using a more complicated expression such as a second-order expression or an expression using a special function.
[0036]
The speech coding apparatus uses the decoded excitation data after pitch filtering by the filter processing unit 203 as an updated adaptive codebook, and performs coding / decoding of the next subframe.
[0037]
In this way, by performing filtering on the decoded excitation data, the pitch structure of the adaptive codebook can be emphasized. In addition, the latest lag pitch filter enhances the latest pitch structure, and higher quality speech can be obtained by using the processed adaptive codebook.
[0038]
Note that the backward filter described above is of the AR type (a type in which a value once filtered becomes a filter state as it is), but the same effect can be obtained as a MA type filter. However, when the long-term correlation is strong (the pitch structure is clear) and the lag is short, the AR type that can emphasize the pitch component is effective.
[0039]
In the above description, the CELP-type coding apparatus is used as an example. However, not only CELP but also past decoded excitation waveforms are used as decoded waveform data (in this embodiment, “decoded excitation data”). Alternatively, even in the case of an encoding device using a decoded speech waveform, the same effect can be obtained by applying the present invention to the same waveform data.
[0040]
The speech decoding apparatus (decoder) has the same configuration as LPC analysis section 101, adaptive codebook 102, stochastic codebook 103, excitation generating section 104, and LPC synthesis section 105 shown in FIG. Each code transmitted from the digitizing device is decoded to obtain a sound source waveform. In this case, the present invention is used in an adaptive codebook updating part.
[0041]
【The invention's effect】
As described above, according to the present invention, in a CELP-type encoding apparatus that stores and updates past decoded excitation data or decoded speech waveforms as decoded excitation data, such as an adaptive codebook, the state update of decoded excitation data is performed. , Filtering is performed after adding and shifting decoded sound source data using the latest decoded waveform. As a result, if the resulting waveform is used as decoded excitation data to be used for the next encoding or decoding, the decoded excitation data can be processed so as to obtain better sound quality. In addition, by using backward filtering from the future to the past, propagation of transmission errors due to the processing can be reduced. In addition, since the backward filtering includes pitch filtering using the decoded lag, decoded excitation data with a clearer pitch structure can be prepared.
[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 functional block diagram showing an internal configuration of an adaptive codebook according to the embodiment. Description]
101 LPC analysis unit 102 Adaptive codebook 103 Stochastic codebook 104 Sound source creation unit 105 LPC synthesis unit 106 Comparison unit 107 Parameter encoding unit 201 Decoded waveform generation unit 202 Decoded waveform insertion unit 203 Filter processing unit 204 Decoded excitation data memory

Claims (5)

Memory-shifting the decoded excitation data, which is an adaptive codebook, from the future to the past and discarding the portion that came out of the memory; storing the local decoded excitation waveform in a vacant portion in the future; and local decoding. Filtering the decoded excitation data after the insertion of the converted excitation waveform, the adaptive codebook updating method. 2. The adaptive codebook updating method according to claim 1, wherein in the step of performing the filtering, backward filtering for filtering from the future to the past is used. 3. The step of performing filtering, wherein pitch filtering is performed by a pitch filter of a backward filter using a lag corresponding to a code of the adaptive codebook and a gain of a source of the decoded adaptive codebook. Update method of adaptive codebook. 4. A speech coding apparatus for updating an adaptive codebook by the method according to claim 1. A speech decoding apparatus for updating an adaptive codebook corresponding to the method according to any one of claims 1 to 3.

Images