Research On The Speech Coding Of MELP And Its Optimization Based On DSP Platform

With the development of information society, communication technology, which represents the development level of the society, has become an important aspect. Definitely, speech communication is one of the most important communications. Speech communication is promoted mostly by speech coding. In the end of 1930's, the theory of PCM and channel vocoder appears. From then on the international and regional standard organizations have established a lot of standards of speech compression coding. These coding standards play a very significant role in the multimedia, network communication and security communication.MELP (Mixed Excitation Linear Prediction) vocoder is widely applied in commercial and military speech communication because of its low-bit-rate, low-delay and low-complexity. The vocoder performs very well in the benign environment. Its performance will degrade rapidly when the background noise increases. In the practical applications, the vocoder will suffer from all kinds of noise. So, this thesis introduces the method to improve the robustness of the vocoder by de-noising.Currently, the implementations of vocoder system are mainly based on three kinds of equipment: (1) PC;(2) ASIC chips;(3) programmable DSP. The last one is most widely utilized. DM642 is the new generation DSP produced by TI Corporation, which possesses high performance and low power cost. At the 600MHz, DM642 gives the processing capability of 4800 MIPS. And it has lots of peripheral interfaces. So, DM642 is the ideal platform for multimedia applications. The MELP vocoder introduced in this thesis is implemented and optimized on the DM642.The thesis at first surveys the development and the current progress of the speech coding, the principle and performance of MELP vocoder. Secondly, the method of de-noising by wavelet is introduced and it is embed into the MELP vocoder as pre-processing. Comparing the synthesis speech qualities of the pre-processed and un-pre-processed, this paper gives out the results of the experiment and the conclusion. After introducing CCS, which is software platform for the optimization, the methods to optimize the MELP vocoder are presented. The methods cover three aspects and an example is introduced to explain them. The final part is conclusion and tasks in the future.