JPH02146100A - Voice encoding device and voice decoding device - Google Patents

Abstract

PURPOSE:To improve the quality of a voice to be decoded and the encoding efficiency by quantizing the voice according to the characteristics of the voice such as a voiced sound, a voiceless sound, and no voice. CONSTITUTION:This device is equipped with a classifying device 1 which analyzes the characteristics of the input voice signal and classifies it by a voiced sound stationary part, a voiced sound transient part, a voiceless sound, and no sound. This classifying device 1 outputs its decision data to switches 10 and 11, a quantizer 4, and a multiplexer 5 according to the characteristics of the voice signal. Then no sound and voiceless sound as to the voice signal are inputted directly to the quantizer 4, the voiced transient part is inputted to the quantizer 4 through a prediction device 3, and the voiced sound stationary part is compressed by a time-base compressor 2 and then inputted to the quantizer 4 through the prediction device 3, so that they are quantized respectively. Consequently, the quantity of the voice to be decoded and the encoding efficiency are improved.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a voice encoding device and a voice decoding device used in digital communications, voice mail, and the like.

Prior Art FIG. 6(a) shows a conventional speech encoding device, and FIG. 6(b) shows a conventional speech decoding device.

In FIG. 6(a), 14 is A by linear PCM.
As shown in FIG. 7, the predictor 15, which calculates the prediction error and short-term/long-term prediction filter coefficients using the /D-converted audio signal, is a code in which a plurality of signal sequences (representative vectors) of a certain length are stored in advance. A quantizer 16 includes a book 15a and a vector quantizer 15b that vector quantizes the prediction error from the predictor 14 and outputs a representative vector number. This is a multiplexer that multiplexes the quantized value and the representative vector number from the quantizer 15.

In FIG. 6(b), 17 is a separator that separates the short-term prediction filter coefficients and long-term prediction filter coefficients from the encoder from the representative vector number, and 18 is the same code as the codebook 15a of the encoder. An inverse quantizer 19 includes a book (not shown) and outputs the representative vector corresponding to the number from the separator 17 as a prediction error; This is a synthesizer that synthesizes audio signals using representative vectors.

Next, the operation of the above conventional example will be explained.

In FIG. 6(a), when an audio signal A/D converted by linear PCM is input, the predictor 14 calculates a short-term prediction filter coefficient in order to remove the correlation between adjacent sample values of the audio signal. Find the short-term prediction error using the short-term prediction filter coefficients.

Furthermore, in order to remove the periodic correlation of the sound source pinch of the audio signal, a long-term prediction filter coefficient is determined using the short-term prediction error, and a prediction error is determined using the long-term prediction filter coefficient.

The quantizer 15 calculates the squared distance between this prediction error signal sequence and each representative vector of the codebook 15a, and outputs the number of the representative vector with the smallest value as a quantized value.

Therefore, the multiplexer 16 sends the audio signal to the decoder as data compressed into short-term prediction filter coefficients, long-term prediction filter coefficients, and representative vector numbers.

In FIG. 6(b), the synthesizer 19 is equipped with a filter having the opposite characteristics to the filter of the predictor 14 of the encoder, and therefore the representative vector is can be decoded into an audio signal.

Problems to be Solved by the Invention However, since the above-mentioned conventional speech encoding devices and speech decoding devices perform the same processing regardless of the characteristics of the speech, such as voiced, unvoiced, or silent, it is difficult to encode speech at a low bit rate. In this case, since encoding is not performed according to the characteristics of the voice, the quality of the decoded voice is not good and there is a problem that the encoding efficiency cannot be improved.

The present invention is intended to solve these problems, and aims to provide a speech encoding device and a speech decoding device that can improve coding efficiency.

Means for Solving the Problems In order to achieve the above object, the speech encoding device of the present invention classifies a speech signal into a voiced sound steady part, a voiced sound transient part, an unvoiced sound, etc., and divides the voiced sound steady part into a temporal compresses on the axis, outputs the prediction error of the voiced sound transient part and the compressed voiced sound stationary part,
It is designed to quantize prediction errors for unvoiced sounds, etc., voiced sound stationary parts, and voiced sound transient parts.

In addition, in order to achieve the above object, the speech decoding device of the present invention dequantizes the quantization values of unvoiced sounds, etc., voiced sound stationary parts, and voiced sound transient parts, and This synthesizes the voiced sound transient part and the voiced sound transient part into an audio signal, and extends the synthesized voiced sound steady part on the time axis.

Effect of the Invention With the above configuration, the present invention can perform quantization according to the characteristics of voice, such as voiced sound, unvoiced sound, and silence, so that the quality of decoded sound is good and the encoding efficiency can be improved. can.

EXAMPLES Hereinafter, the present invention will be described in detail with reference to the drawings. FIG. 1(a) is a block diagram showing an embodiment of a speech encoding device according to the present invention, FIG. 1(b) is a block diagram showing an embodiment of a speech decoding device according to the present invention, 2 is a detailed block diagram showing the predictor in FIG. 1(a), FIG. 3 is a detailed block diagram showing the quantizer in FIG. 1(a), and FIG. 4 is a detailed block diagram showing the quantizer in FIG. 1(a). The waveform diagram of the audio signal, Figure 5, is similar to Figure 1 (a
) is a flowchart for explaining the operation of the classifier.

In FIG. 1(a), 1 is a classifier that analyzes the characteristics of the input audio signal and classifies it into a voiced stationary part, a voiced transient part, an unvoiced sound, and a silent part, as will be described later; and 2 is a classifier. This is a time axis compressor that thins out similar waveforms in the voiced stationary part of the audio signal and compresses them on the time axis based on the judgment data from No. 2.

3, due to the voiced sound of the audio signal, its prediction error and short-term/
Predictor 3 is a predictor that calculates long-term prediction filter coefficients.
As shown in FIG. 2, a short-term prediction filter 21 is used to obtain a short-term prediction error based on voiced sounds, and a short-term prediction filter 2 is used to remove correlations between adjacent sample values of voiced sounds.
short-term prediction analyzer 22 for determining short-term prediction filter coefficients of 1;
and a long-term prediction filter 23 which calculates a prediction error using the short-term prediction error from the short-term prediction filter 21; a long-term prediction analyzer (pitch predictor) 24 that calculates prediction filter coefficients;
It is comprised of a quantizer 25 that quantizes short-term predictive filter coefficients and long-term predictive filter coefficients, respectively.

4 is a quantizer that vector-quantizes the prediction error from the predictor 3, and the quantizer 4 is used for voiced sound stationary parts, voiced sound transient parts, and unvoiced sounds, as shown in FIG. , codebook 3 in which representative vectors for silence are stored, respectively.
1 to 34, and one of the codebooks 31 to 34 is selected based on the judgment data of the audio signal classified by the classifier 1,
It is composed of a vector quantizer 35 that calculates the squared distance between the prediction error signal sequence from the predictor 3 and each representative vector of the codebook, and outputs the number of the representative vector with the smallest value as a quantized value. ing.

5 is a multiplexing unit that multiplexes the judgment data of the audio signal classified by the classifier 1, the quantized value of the short-term/long-term prediction filter coefficient from the predictor 3, and the number of the representative vector from the quantizer 4. 10 and 11 are switches that are switched according to the characteristics of the voice classified by the classifier 1, respectively.

In FIG. 1(b), a separator 7 separates transmission data from the encoding device into voice signal judgment data, short-term/long-term predictive filter coefficient quantization values, and representative vector numbers;
is equipped with the same codebook (not shown) as codebook 7 31 to 34 of the encoder, selects the codebook based on the judgment data of the audio signal from the separator 6, and selects the codebook corresponding to the number from the separator 6. an inverse quantizer 8 that outputs a representative vector as a prediction error; a synthesizer 9 that synthesizes a voiced sound of the audio signal using the short-term predictive filter coefficients, long-term predictive filter coefficients, and the representative vector from the inverse quantizer 8; 12 and 13 are switches that are switched according to the judgment data of the audio signal from the separator 6, respectively. The operation of the embodiment will be explained.

In FIG. 1(a), an audio signal that has been A/D converted by linear PCM is input to an encoding device every fixed time length (frame).

As shown in FIG. 4, the audio signal (analog signal) includes silence, a relatively long unvoiced sound that is the beginning of audio, and a relatively short voiced transition part that continues between the unvoiced sound and the voiced sound steady part.
It is composed of voiced stationary parts.

In this audio signal, (1) the voiced sound stationary part is characterized by a series of similar waveforms that exhibit periodicity, and the correlation between samples and pitch periods is strong, and (2) the voiced sound transient part is , similar waveforms are not continuous, but have the characteristic of strong correlation between samples and during the pinch cycle; (3) unvoiced sounds have the characteristic of weak correlation between samples and during the pitch cycle, and the signal changes rapidly; (4) Silence has a characteristic that the amplitude of the signal is small.

When such an audio signal is input, the classifier 1 calculates the power Pw of the waveform within one frame, as shown in FIG. 52). If the power Pw is less than the threshold, the frame is determined to be silent (step 53);
If it is equal to or greater than the threshold, it is determined that there is a sound, and the process proceeds to step 54 and subsequent steps.

In step 55, the autocorrelation coefficient r(i) of the waveform within one frame is calculated, and the maximum value rmax at the specified i is determined from the calculated autocorrelation coefficient r(i) (step 56). It is determined whether the maximum value rmax is greater than or equal to the threshold (step 57). If the maximum value rmax is less than the threshold, the frame is determined to be an unvoiced sound part (step 58), and if it is greater than the threshold, it is determined to be a voiced sound and the process proceeds to step 59 and subsequent steps.

In step 60, it is determined whether or not the previous frame is a voiced sound. If NO, the frame is determined to be a voiced sound transition part (step 61); if YES, the process proceeds to step 62 and subsequent steps.

In step 62, the pitch period Pn is calculated using the autocorrelation coefficient r (i), and the pitch period Pn- of the previous frame is calculated.
The rate of change ρ from 1 is calculated (step 63), and it is determined whether the rate of change ρ is greater than or equal to a threshold (step 64). Rate of change ρ
If is less than the threshold, the frame is determined to be a voiced sound stationary part, and if it is greater than or equal to the threshold, the frame is determined to be a voiced sound transient part.

The classifier 1 outputs, for example, 2-bit judgment data to the switch 10.11, the quantizer 4, and the multiplexer 5, depending on the four types of characteristics of the audio signal.

In a frame determined to be a voiced sound stationary part, the switch 10 is switched to the time axis compressor 2 side, and in a frame determined to be a voiced sound, the switch 11 is switched to the predictor 3 side.

Therefore, silence and unvoiced sounds in the audio signal are directly processed by the quantizer 4.
and are quantized into representative vector numbers of the silence codebook 34 and the unvoiced sound codebook 33, respectively.
The voiced sound transient part is inputted to the quantizer 4 via the predictor 3 and quantized into the representative vector number of the codebook 32 for voiced sound transient parts, and the voiced sound steady part is compressed by the time axis compressor 2. After that, it is input to the quantizer 4 via the predictor 3 and quantized into the number of the representative vector of the codebook 310 for voiced sound stationary part.

Here, when encoding silence and unvoiced sounds, the predictor 3
Since the bits assigned to the predictive filter coefficients etc. are assigned to the focus of the quantizer 4, the number of transmitted bits is kept constant.

The judgment data of the speech signal is transmitted to the decoder along with the predictive filter coefficients and the number of the representative vector, so the decoding device can decode it to the original speech signal, and also distinguish between voiced, unvoiced, silent, etc. Since the quantized value is decoded according to the quantized value, the quality of the decoded voice is good.

As described in detail, the speech encoding device of the present invention classifies an audio signal into a voiced sound steady part, a voiced sound transient part, an unvoiced sound, etc., compresses the voiced sound steady part on the time axis, a voiced transition part,
The speech decoding device of the present invention outputs the prediction error of the compressed voiced sound stationary part and quantizes the prediction error of the voiced sound steady part and the voiced sound transient part. Then, the quantized values of the voiced sound steady part and the voiced sound transient part are dequantized, and the dequantized voiced sound steady part and voiced sound transient part are synthesized into a speech signal, and this synthesized voiced sound steady part is Since it is expanded on the time axis, it is possible to quantize according to the characteristics of the voice, such as voiced, unvoiced, or silent, which improves the quality of the decoded voice and improves the encoding efficiency. can be improved.

[Brief explanation of the drawing]

FIG. 1(a) is a block diagram showing an embodiment of a speech encoding device according to the present invention, FIG. 1(b) is a block diagram showing an embodiment of a speech decoding device according to the present invention, Figure 2 shows
FIG. 3 is a detailed block diagram showing the predictor of FIG. 1(a); FIG. 3 is a detailed block diagram showing the quantizer of FIG. 1(a);
Figure 4 is a waveform diagram of a general audio signal, and Figure 5 is a waveform diagram of a general audio signal.
FIG. 6(a) is a flowchart explaining the operation of the classifier in FIG. 6(a). FIG. 6(b) is a block diagram showing a conventional speech encoding device. The block diagram shown in FIG. 7 is a detailed block diagram showing the quantizer of FIG. 6(a). 1... Classifier, 2... Time axis compressor, 3... Predictor, 4... Quantizer, 5... Multiplexer, 6... Separator,
7... Inverse quantizer, 8... Synthesizer, 9... Time axis expander, 31... Code book for voiced sound stationary part, 32...
・Code book for voiced sound transition part, 33... Code book for unvoiced sound, 34... Code book for silent sound. Agent's name: Patent attorney Shigetaka Awano and 1 other person ■ 4N1 Rhinoceros ■ Hanawa P Lukkou County ■ ν Shield■ Netsushishi Diagram

Claims (3)

[Claims] (1) means for classifying the audio signal into a voiced sound steady part, a voiced sound transient part, an unvoiced sound, etc., and a means for compressing the voiced sound steady part of the sound signal classified by the classification means on the time axis;
means for outputting a prediction error of the voiced sound transient part of the audio signal classified by the classification means, the voiced sound steady part compressed by the compression means, the unvoiced sound, etc. of the audio signal classified by the classification means; A speech encoding device comprising: means for quantizing prediction errors of the voiced sound stationary part and the voiced sound transient part. (2) The quantization means includes a codebook in which representative vectors for a voiced sound stationary part, a voiced sound transient part, an unvoiced sound, etc. are stored, and converts the voiced sound steady part, voiced sound transient part, unvoiced sound, etc. into the code 2. The speech encoding apparatus according to claim 1, wherein the speech encoding apparatus performs vector quantization using a book and outputs a number of a representative vector. (3) Means for dequantizing the quantization values of unvoiced sounds, etc., voiced sound steady parts and voiced sound transient parts, and means for synthesizing the dequantized voiced sound steady parts and voiced sound transient parts into an audio signal. and means for expanding the synthesized voiced sound stationary part on a time axis.