Research On Speech Algorithm Of MELP And Realization On DSP

ABSTRACT:Voice is one primary means of exchanging information,digitized voiced signals for transmission, storage and recognition constitute the most important and basic components in current telecommunications industry, In the wired communications, with the fast growing popularity of fiber-optic technology, channel bandwidth and transmission capacity can be guaranteed greatly. However, in wireless communications, such as voice communication systems, wireless cha nnel bandwidth of the channel is always a bottleneck. In digital mobile communic ations and multimedia communications,especially voice communication systems, such as the rapidly developing digital radio, there is an urgent need to study a cod-ing schemes that can both save channel bandwidth and transmit the information efficiently and reliably, especially those vocoders with low rate speech coding rate, high-quality decoding voice.Mixed excitation linear prediction (MELP) voice compression codec algorithm is linear predictive coding (LPC) on the basis of the model parameters, using the mixed excitation,and the multi-band ideas. Lower coding rate and high-quality decoding voice makes MELP speech coding algorithm the ideal encoding scheme in low rate speech coding algorithm.This paper analyzes how the MELP speech coding algorithm works, and study depth in its key technologies,encoding and decoding process. And it has a detailed argument on the possibility that MELP may be used in digital radio platform.The realization of MELP algorithm on TI's high-performance floating-point proc-essor TMS320C6722B is the key of the paper. C6722B integrats lots of interfaces inside the chip for voice applications. In order to meet the voice codec real-time requirements and make it work well, the paper has done a lot of optimization wo-rk, and write the drivers for the corresponding on-chip peripheral.Experiments show that the C6722B system can run the MELP speech codec algo-rithm efficiently and reliably; real-time sampled voice signal goes through with the MELP coding and decoding process,and is played out, we can get clear and natural audio effects. Make micro-analysis on decode speech we conclued that the decoded voice waveform and the original speech waveform is very similar, there is attenuation and delay only in large energy range of the signal.