Multi-description speech coding and decoding methods and systems based on linear predictive residual classified quantization

Abstract

Multi-description speech coding and decoding methods and systems based on linear predictive residual classified quantization are provided. Short-time prediction parameters, long-time prediction parameters and a time-domain predictive residual signal are obtained through short-time prediction analysis and long-time prediction analysis of a to-be-coded voice stream, the residual signal is divided into multiple multidimensional coding vectors, the energy envelope of each coding vector is calculated for bit allocation, and dual-cycle vector quantizing and coding is carried out on the time-domain coefficient of each coding vector; and interlaced multi-description grouped packaging is carried out on the short-time prediction parameters and the long-time prediction parameters, and multiplexed packaging is carried out on the predictive residual quantization indexes and the coded bits of the vector quantization values of quantized coded signals.

Description

technical field [0001] The invention relates to the field of speech codec and transmission, in particular to a multi-description speech coding and decoding method and system based on linear prediction residual classification and quantization. Background technique [0002] In many application scenarios, voice communication has high requirements for real-time and continuity, but the current network often loses voice data arriving at the receiving end due to network delay, network congestion, and transmission errors, resulting in a serious decline in voice quality; On the other hand, in order to obtain better communication voice quality, it is required that the compressed voice signal should have higher definition. Therefore, for VoIP, the problem of coded voice quality and data loss compensation of unreliable packet networks are contradictory communities. The traditional method to deal with packet loss is retransmission, but when the packet loss rate is high, retransmission w...

AI Technical Summary

Problems solved by technology

The multi-descriptive waveform speech coder proposed earlier has a simple algorithm, which can improve the transmission stability of the system, but the compression rate of this type of coder is not high, which affects the quality of coded speech

Subsequent code-excited linear prediction (CELP) speech multi-description encoders, these encoders have high enough compression efficiency, but there is a strong dependence between their parameters, the description decomposition method is not flexible enough, and the improvement of stability is At the cost of greatly reduced performance, at the same time, because CELP coding adopts adaptive codebook and random codebook codebook excitation methods for speech coding, its coding quality reaches a limit at a certain medium and low bit rate, and cannot further improve the normal Speech codec quality under non-drop frame conditions

Images