JPS594718B2 - Audio encoding method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a speech encoding method that predictively encodes an audio signal by adaptively changing its encoding characteristics depending on its properties.

Such a predictive coding method was proposed in Japanese Patent Application No. 1983-42858 entitled "Adaptive Predictive Coding Method".

That is, in this method, an audio signal is divided into a plurality of bands, prediction coefficients are determined for each band, and the signals in each band are predictively encoded using the prediction coefficients. At that time, the amount of transmitted information, that is, the number of quantization levels, is allocated and encoded temporally non-uniformly according to the power ratio between the signals of each divided frequency band and the temporal locality of the predicted residual power. 5. Make sure that the quantization error before and after is small. In this way, even if the transmission capacity is relatively small, a good quality and therefore naturalness is obtained. However, in this conventional adaptive bit allocation predictive coding method, a prediction residual 10 difference signal is sent for each divided frequency band, and the amount of information in this prediction residual signal is relatively large, and the overall amount of transmitted information is large. There was a drawback.

It is an object of the present invention to provide a speech encoding system that has good quality and naturalness with a small amount of transmitted information15.

According to this invention, an audio signal is divided into a plurality of frequency bands, some of the lower bands are coded using adaptive predictive coding, and only prediction coefficients representing spectral characteristics are used for the upper bands. The code is changed to 0 and sent.

In decoding, the low frequency component is obtained by decoding the adaptive prediction code, and the high frequency component is obtained by creating a pseudo residual signal from the low frequency band and decoding it from this and the prediction coefficient. FIG. 1 shows a fifth embodiment of the speech encoding system according to the present invention.

The digitized audio signal from the input terminal 1 is divided into a plurality of frequency bands, for example, Li, L2, and H, as shown in FIG. The low ? width is resampled at the Nyquist frequency as shown in Figure 2B. It is converted to a 0 band signal. For each of the low-frequency signals Li', L2', and Raccoon, the linear prediction analysis unit 3 extracts prediction coefficients for each band, and the residual power calculation unit 4 calculates the residual power divided into two dimensions by band and time. Extracted.
Low frequency cha 95 in the audio signal in the low frequency signal
NnelL,',L. ' is predictively coded by the predictive coding unit 6. In the predictive coding unit 6, the predictive coefficient, quantization width, and number of quantization levels are adaptively changed according to the characteristics of the low-frequency signal for each short time. Ru. The number of quantization levels is allocated by the bit allocation unit 5 in terms of time and frequency under the condition that the information rate is constant according to the deviation of the residual power between bands and the deviation in time within the band. It will be done. These specific methods are shown in the specification of Japanese Patent Application No. 54-42858. For the residual signal and low frequency channels L/, L2', the transmission path encoding unit 7 processes the residual signal (predicted For the high frequency channel w of the audio signal in the low frequency signal, the output of the encoding unit) is combined with the prediction coefficient and the average residual power parameter in each frame, and is converted into a bit sequence and sent to the transmission path 8. Send.

On the decoding side, a transmission path decoding unit 9 separates the residual signal and parameter information and decodes the parameter information.

The bit allocation unit 10 calculates the quantization width and the number of quantization levels from the residual power as shown in the patent application specification, and using this, the residual signal decoding unit 11 calculates the residual signal of the low frequency channel. SL1 and SL2 (FIG. 2C) are decoded and converted into low frequency signals in the predictive decoding section 12. On the other hand,
The above-mentioned residual signals SLl and SL2 are processed in the band signal converter 13 by dividing the band residual signals SL/, SL2' (FIG. 2D) having the same sampling frequency as the input signal, that is, the audio signal into a plurality of bands. The pseudo residual signal S is converted into a corresponding frequency band before being converted to a low frequency band, and is converted into a low frequency signal band in the pseudo residual signal generation unit 14, and is converted into a pseudo residual signal S having high frequency components.
n'' (Fig. 2E) is obtained.

The pseudo residual signal generation unit 14 generates band residual signals SL/,SL
As shown in Figure 3A, the sample values of 2'' are separated by several sample value intervals (4 sample value intervals in this figure), leaving only those indicated by arrows.
The other sample values are set to zero. In that case,
It is preferable to reset the interval at which sample values are left as shown in FIG. 3B at a timing corresponding to the pitch period. In FIG. 3B, the last arrow indicates three sample values. In addition,
The pitch period can be calculated by the peak search of the autocorrelation coefficient of the band residual signals SL/, SL2'' in the pitch period calculating section 15. is converted into a low-frequency residual signal for channels H, and converted into a low-frequency signal in the predictive decoding section 12.In the band signal conversion section 17, the predictively decoded low-frequency channels L/, L2' and From the low frequency signal of each band of the high frequency channel F, the band signal Ll for each band,
The output audio signal is obtained by converting into L2 and H and adding these. As explained above, this audio encoding method uses highly efficient waveform encoding in the low frequency range and generates a pseudo residual signal on the receiving side without transmitting the residual signal in the high frequency range.
The amount of transmitted information can be significantly reduced.

In particular, if the pseudo residual signal is generated as shown in FIG. 3B, the harmonic structure is maintained and high quality speech can be provided at an information rate of 6.4 to 9.6 kilobits/second. As a specific example, the audio band is O~800Hz, 8
The voice quality when the low frequency band 0 to 1600 Hz is divided into three bands, 0 to 1600 Hz and 1600 to 3200 Hz is adaptive bit allocation predictive coding, and 1600 to 3200 Hz is spectrum coded, the entire band is adaptive bit allocation predictive coding. It was confirmed that the results were superior in terms of satisfaction and brightness compared to the case where the

Furthermore, compared to the case where a pulse signal is used to generate a pseudo residual signal, generating a pseudo residual signal using the method shown in FIG. 3B is more effective in not deteriorating the clarity of speech. confirmed.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the audio encoding method according to the present invention, FIG. 2 is a diagram for explaining its operation, and FIG. 3 is a diagram for explaining the generation of a pseudo residual signal. be. 1: Signal input terminal, 2: Low frequency signal conversion section, 3: Linear prediction analysis section, 4: Residual power calculation section, 5: Bit allocation section, 6
: Predictive coding unit, 7: Transmission line coding unit, 8: Transmission line, 9
: Transmission path decoding unit, 10: Bit allocation unit, 11: Residual signal decoding unit, 12: Predictive decoding unit, 13: Band signal conversion unit, 14: Pseudo residual signal generation unit, 15: Low frequency signal Conversion unit, 16: Pitch extract, 17: Band signal conversion unit, 18
: Signal output terminal.

Claims (1)

[Claims] 1 The audio signal is divided into a low frequency band and a high frequency band, and the signal in the low frequency band is encoded by adaptive bit allocation predictive coding, and the signal in the high frequency band is encoded by
A speech encoding method that obtains and encodes prediction coefficients representing spectral characteristics, and outputs prediction codes for these low-frequency band signals and codes for prediction coefficients for high-frequency band signals.