Mflp And Very Low Rate Speech Codec Algorithm Research

With the development of digital communication business, and with the urgent demands for very low bit rate vocoders in bad wireless communication environments, vocoder algorithms that can rebuild good quality speech at 2400 bits per second or even lower bit rate has become a hot research.Mixed excitation linear prediction (MELP) vocoder, is one of the most potential algorithm in modern low bit rate speech coding. This algorithm introduced several improvements to traditional LPC Vocoder, so that this vocoder can simulate more natural speech characteristics and can achieve high synthesis speech quality.With MELP as the fundermental model, this paper studied speech coding parameters such as: pitch, gain, LSF, Unvoiced/Voiced and excitation spectrum in detail, with emphasis on their relativity between frames, statistical features, and their influence to descibe natural speech. The choice, extraction and quantization methods of the parameters in very low bit rate speech coders are discussed to design the desired vocoder.Very low bit rate speech coders can hardly get good performance, for there are not sufficient bits to present spectrum of speech signal. High efficient quantization, multi-frame joint, parameter interpolate and frame length enlargement techniques are promoted in this paper to reduce speech coding bit rate as well as guarantee speech quality. With these techniques, two very low bit speech coders at 1.2kbps and 0.8kbps are designed and implemented in this subject, principles of the two vocoders are described, their performance for different languages in diverse channels is also evaluated in the paper.Informal subjective listening tests show that the speech quality of 1.2kbps vocoder is better than that of traditional LPC 2.4kbps algorithm, and is comparable with American Federal Standard 2.4kbps MELP. The synthesized speech of 0.8kbps vocoder achieves high intelligibility, clearness, and relative naturalness.