Study Of Efficient Vocoder Algorithm Based On MELP

Recently, with the development of broadband communication, it seems that the band is not a serious problem any more in the field of line communications. But in wireless communication field such as mobile communications and satellite communications band is always a kind of rare resource. On the one hand multi-channel coded voice can be transmited in channel with limited bandwidth, to improve utilization ratio of transmission channel; in the other, digital signals can be transmited in the analog channel with narrowband (HF), especially in the military communications systems require sophisticated encryption of data, speech coding play an irreplaceable role. In recent years, applications, such as various digital voice storage and voice response, also make an urgent request to speech coding algorithm that can synthesis voice with high quality.MELP Vocoder is a perfect coding scheme in low rating voice coding relatively. The MELP Algorithm is based on the model of LPC and use the form of mixed excitation and MBE, better simulate the characteristics of natural language, synthesis high-quality speech in such a relatively low rate. But with high computation complexity of MELP Vocoder, resulting in relatively difficult for VLSI implementation. In this paper, we mainly study the theory of MELP, Including the procedure of coding and decoding , and then make some improvements to the algorithm. Base on depth-in study on MELP Algorithm, we found mainly these bottlenecks: exponential computation, fourier series amplitude computation, pitch detection, LSF quantization. How to reduce the complexity and difficulty of VLSI implementation become the major content of this paper. In this paper, we use a look-up table that can be used for exponential computation and quantification, under guarantee the computing precision, this method can reduce the difficulty of VLSI implemention for exponential component. we use a relatively low complexity of heuristic pitch detection algorithm, greatly increase the efficiency of pitch detection, and reduce difficulty of VLSI implementation; a new method based on the quantification of regional has be used for LSF quantification, greatly improve the efficiency of the algorithm, and greatly reduce the complexity of VLSI implementation. In the end MELP and Improved MELP is implemented using C programming language.Results show that the improved algorithm greatly reduce the computation complexity, but with good systhesised speech. Finally, conclusions and future work of this paper is presented.