JPH04270397A - Voice encoding system - Google Patents

Abstract

PURPOSE:To improve the tone quality by obtaining a pattern of a smooth delay of an adaptive code book, and also, suppressing a noise sense which is a problem point of a voice encoding system of a CELP system, in the voice encoding system of about 8-4kb/s. CONSTITUTION:Valcalization is discriminated at every frame, and in the frame which is discriminated to be valcalization, at the time of searching an adaptive code book, to a sound source for delaying a sound source in the past, a COM filter corresponding to its delay is set, and a delay by which a square error of an output of the COM filter and a weighting signal becomes minimum is derived. In the case it is not discriminated to be valcalization, the COM filter is not used. At the time of being voicedness, with respect to an output of the adaptive code book, two ways of constitutions in the case the COM filter is used, and in the case the filter is not used can be taken.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an audio encoding method for encoding audio signals with high quality at a low bit rate, particularly about 8 to 4 kb/s.

0002

2. Description of the Related Art For example, M. S
Chroeder and B. S. “Code-excited linear” by Atal
Prediction：High quality
speech at low bit ra
tes” (Proc. ICASSP, pp.
．． 937-940, 1985) (Reference 1), etc.
PC Coding) is known. In this method, on the transmitting side, the spectral parameters representing the spectral characteristics of the audio signal are extracted from the audio signal every frame (for example, 20 ms), the frame is further divided into subframes of small intervals, and the past audio source signal is Extract the pitch parameter representing the long interval correlation (pitch correlation) from
Long-term prediction of subframe audio signals is performed using pitch parameters. For the residual signal obtained by long-term prediction,
Select one type of noise signal to minimize the residual signal and error power with a signal synthesized with a signal selected from a codebook of noise signal types prepared in advance, and calculate the optimal gain. . Then, the index and gain indicating the type of the selected noise signal, as well as the spectrum parameter and pitch parameter are transmitted. A description of the receiving side will be omitted.

[0003] Furthermore, as a method of long-term prediction, for example,
W. Kleijn “An efficient sto
chastically excited lin
early predictive coding al
gorithm for high quality
low bit rate transmi
The paper entitled “Ssion of speech” (Sp
eech Communication, 7, pp.
305-316, 1988) (Reference 2), the past sound source is shifted one sample at a time, and the sound source is shifted with the shift value (integer delay) of the past sound source that minimizes the weighted squared error. There is a known method for finding the gain of a sound source corresponding to . However, since the adaptive codebook alone cannot completely describe the periodicity of speech, CEL
In the case of the P-based encoding system, a sense of noise remains in the reproduced signal. To solve this problem, a known method is to multiply the sum of the adaptive codebook and the excitation codebook, which are input to the synthesis filter, by an adaptive comb filter defined by the delay of the adaptive codebook. For example, the specific method is
S. “Improved exc.
itation for phoneticall
y-segmented VXC speech
Coding below 4kb/s” (Proc. Globecom, pp.946-
950, 1990) (Reference 3). With this method, when the delay matches the pitch period of the audio signal and stops correctly, the feeling of noise is reduced and the sound quality is improved.

0004

[Problems to be Solved by the Invention] However, adaptive comb filtering has a problem in that if the delay is not properly controlled, the sound quality will deteriorate on the contrary.

An object of the present invention is to solve the above-mentioned problems and to provide a speech encoding system with good sound quality at 8 to 4 kb/s using an adaptive codebook sound source with smooth delay and harmonic components added. Our goal is to provide the following.

[0006]

[Means for Solving the Problems] The audio encoding method according to the first invention includes means for accumulating an audio signal, means for determining voicing of the audio signal, and means for dividing the audio signal into subframes. means for analyzing a speech signal; means for applying perceptual weighting to the speech signal; means for performing comb filtering on the output of an adaptive codebook only for frames determined to be voiced; and a comb filter. means for determining the optimal delay using the output of the adaptive codebook, means for subtracting the output of the adaptive codebook corresponding to the optimal delay from the weighted signal, and means for extracting the optimal sound source from the sound source codebook. shall be.

[0007] The audio encoding method in the second invention includes means for accumulating an audio signal, means for determining voicing of the audio signal, means for dividing the audio signal into subframes, and means for analyzing the audio signal. , a means for adding perceptual weighting to the speech signal, a means for performing comb filtering on the output of the adaptive codebook only for frames determined to be voiced, defined by the delay, and an optimal method using the output of the comb filter. The present invention is characterized by comprising means for determining a suitable delay, means for subtracting the output of a comb filter corresponding to the optimal delay from the weighted signal, and means for extracting an optimal sound source from a sound source codebook.

[0008]

[Operation] In the first invention, instead of using the conventional weighted squared distance as the evaluation measure, a comb filter defined by the delay is applied to the output of the adaptive codebook to emphasize harmonic components. The squared distance between the output and the weighted signal is used as an evaluation measure, and the delay that minimizes this distance is determined as the optimal delay. By applying a comb filter, not only harmonic components are emphasized, but also noise components are suppressed, so that a more accurate delay can be estimated.

In the second invention, by adding a comb filter to the evaluation scale, it is possible to estimate a more accurate delay, and by using the output of the comb filter as the output of the adaptive codebook, CELP-based encoding It is possible to reduce the feeling of noise, which was a problem with this method.

0010

Embodiment FIG. 1 shows an embodiment of the first invention, and FIG. 2 shows an embodiment of the second invention.

First, the operation of each module will be explained.

Buffer device 105 is a device that stores audio signals.

The voicedness determiner 110 is a device that determines whether or not the voice of the frame is voiced.

The subframe divider 120 is a device that divides the audio signal stored in the buffer into several subframes.

The LPC analyzer 200 is a device that extracts LPC coefficients, which are voice spectral parameters, for each frame.

The above buffer device 105 and audio determiner 1
10, subframe divider 120, LPC analyzer 200
Use the existing one.

The LPC coefficient quantizer 210 is a device that quantizes LPC coefficients, and any method may be used.

Weighting filter 130 performs well-known perceptual weighting on the audio signal divided into subframes.

The influence signal subtractor 140 subtracts the influence signal calculated by zero-exciting the weighted synthesis filter with the initial state set to the last state of the weighted synthesis signal one subframe before, from the weighted signal. .

The adaptive codebooker 150 generates a sound source that is a delayed version of a past sound source. The delay may be an integer or a decimal number, but a decimal number provides better sound quality. In the case of a decimal number, the past sound source is reduced to a decimal number by polyphase filtering. Details of polyphase filtering can be found in, for example, P. Kroon “Pitch predictor
s with high temporal
A paper titled “Proc.
ASSP, pp. 661-664, 1990) (Reference 4).

The comb filter 160 performs comb filtering on the delayed sound source of the past sound source generated by the adaptive codebook unit 150 in accordance with the delay. The shape of the comb filter may be either IIR type or FIR type. For the specific method, reference can be made to the above-mentioned document 3 and the like.

[0022] The weighting filter 170 synthesizes the LPC coefficients obtained through the analysis using well-known perceptually weighted filter coefficients.

The delay determiner 180 is a circuit that determines the delay that minimizes the squared distance between the comb-filtered past delayed sound source and the weighted signal.

A sound source codebook searcher 190 searches for a sound source codebook. In this case, the sound source codebook may be a noise codebook or a learning codebook learned by a VQ algorithm such as the LBG method. For the former, reference can be made to the above-mentioned document 1. Also, the latter is
-42955 specification (Document 5), patent application No. 2-429
Specification No. 56 (Document 6) can be referred to. 220 is a multiplexer.

First, the operation of the embodiment shown in FIG. 1 will be explained.

A voice signal is input from the signal line 100 and stored in a buffer 105, and a voiced determiner 110 determines vowels using the stored signal. An LPC analyzer 200 performs LPC analysis to calculate LPC coefficients, which are spectral parameters. The calculated LPC coefficient is LP
The C coefficient quantizer 210 quantizes the signal and sends it to the multiplexer, where it is again decoded into LPC coefficients and used in the following processing. The stored audio signal is divided by the subframe divider 120, and the following processing is performed on the signal for each subframe. First, the weighting filter 130 applies perceptual weighting to the audio signal, and the influence signal subtracter 140
Now calculate the influence signal and subtract it from the weighted signal. An adaptive codebook unit 150 generates a sound source signal by delaying a past sound source, and a comb filter 160 performs comb filtering in accordance with the delay value for frames determined to be voiced by the voicing determiner 110. If it is determined that the voice is not voiced, comb filtering is not performed. Next, the output of the comb filter 160 is input, and weighted synthesis is performed. The delay determiner 180 determines the delay that minimizes the square distance between the weighted composite signal and the weighted signal. After the optimal delay is determined, the output of the adaptive codebook corresponding to the optimal delay is input, weighted combination is performed, and the weighted combined signal by the adaptive codebook is subtracted from the weighted signal. For the subtracted signal, a sound source codebook searcher 19
0 to search the sound source codebook. multiplexer 2
20 combines and outputs the output code sequences of the LPC coefficient quantizer 210, the adaptive codebook delay determiner 180, and the excitation codebook searcher 190. These processes are performed for each subframe.

Next, the operation of the embodiment shown in FIG. 2 will be explained.

In the first invention, the output of the comb filter 160 is used only for determining the delay, but in the second invention, the only difference is that the output of the comb filter 160 is used as the output of the adaptive codebook. There is. here,
Only that point will be explained. After the delay is determined by the delay determiner 180, the output of the comb filter 160 is used as the output of the adaptive codebook, and the synthesis filter 170 performs weighted synthesis using it as input, and then subtracts it from the weighted signal to obtain the excitation code. A book searcher 190 searches the sound source codebook.

This completes the description of the embodiments of the present invention.

Various modifications other than the configuration of this embodiment are possible.

In calculating the delay of the adaptive codebook, in order to reduce the amount of calculation, multiple types of delay candidates are obtained by correlation calculation using input weighted signals, etc., and for each candidate, a range of neighboring samples is determined. It is also possible to adopt a configuration in which the optimal decimal delay is determined by .

In addition, instead of uniquely determining the adaptive codebook and the excitation codebook in each subframe, multiple types of candidates are determined in descending order of error power, and these are accumulated over the entire frame to determine the cumulative error power over the entire frame. A combination of an adaptive codebook and a sound source codebook that minimizes the above may be selected.

[0033]

[Effects of the Invention] As described above, according to the present invention, by using an adaptive comb filter in an adaptive codebook in a voiced section of speech, a smooth delay can be obtained and the delay in the adaptive codebook can be reduced. This has the great effect of providing good reproduced audio with less noise even when a mistake is made.

[Brief explanation of the drawing]

FIG. 1 is a configuration diagram showing a first embodiment of the present invention.

FIG. 2 is a configuration diagram showing an embodiment of the second invention.

[Explanation of symbols]

100 Audio input port 105 Buffer device 110 Voiced determiner 120 Subframe divider 130 Weighting filter 140 Influence signal subtractor 150 Adaptive codebook device 160 Comb filter 170 Weighted synthesis filter 180 Delay determiner 190 Sound source codebook searcher 200 LPC analyzer 210 LPC coefficient quantizer 220 Multiplexer

Claims (2)

[Claims] Claims 1: means for accumulating an audio signal; means for determining voicing of the audio signal; means for dividing the audio signal into subframes; means for analyzing the audio signal; means for applying weighting to the output of the adaptive codebook only for frames determined to be voiced; means for determining an optimal delay using the output of the comb filter; 1. A speech encoding method comprising: means for subtracting an output of an adaptive codebook corresponding to an optimal delay from a signal; and means for extracting an optimal sound source from the sound source codebook. 2. Means for accumulating a voice signal, means for determining voicing of the voice signal, means for dividing the voice signal into subframes, means for analyzing the voice signal, and means for determining the voicing of the voice signal. means for applying weighting to the output of the adaptive codebook only for frames determined to be voiced; means for determining an optimal delay using the output of the comb filter; 1. A speech encoding method comprising: means for subtracting the output of a comb filter corresponding to a delay from a weighted signal; and means for extracting an optimal sound source from a sound source codebook.