Research And Implemention Of Low-Rate Speech Coding Based On MELP

Speech is an important way for people to communicate and also is one of the most significant information media. Speech compression plays an important part in storage and digital communication of speech signal because of its redundancy. With the rapid development of communication and computer network technology, speech coding has wide application in the area such as mobile communication, satellite communication, multimedia technology and IP phone .But for the purpose of saving bandwidth in some application area, reducing speech coding rate is urgent. So low-rate speech coding is developed increasingly and has a wide application in military communication.Parameter coding is the main method of low-rate speech coding and multi-band mixed excitation linear prediction (MELP) coding is one of important parameter coding. Department of Defense Digital Voice Processing Consortium(DDVPC) has selected MELP vocoder at 2.4Kbps as new Federal standard. In this paper, several research and many experiments have been carried out on the aspects of speech analysis, parameter coding and speech synthesis based on the standard algorithm. LPC (Linear Predictive Coding) is one of the most important parameters in MELP speech model. LSF (Line Spectrum Frequency) is another expression form of LPC .It make up the inadequate of all-pole model. This paper makes research on linear prediction and analyses anti-noise performance of LPC coefficients.In order to reduce the rate of speech coding, this paper presents new 1.2Kbps MELP algorithm using sufficient vector quantization based on 2.4Kbps MELP standard algorithm and implement by C language. The key point of new algorithm is joint-parameter quantization. According to simulation results carried out on kinds of different codebook design algorithm, full-search cookbook designed using LBG algorithm is sufficient. It can reduce coding rate and gain better synthesized speech.