Research On MELP Algorithm And Realization Of Embedded Platform

Speech is one of the most basic communication means in human communication. In order to meet the needs of saving bandwidth and reducing speech coding rate, in modern speech communication system, low rate speech coding technology has been a hot research topic. It is widely used in satellite communication, military applications and secure communication, etc.Low rate speech coding technology has high complexity and complicated computation, but it also has very high practical value. MELP algorithm is relatively better among the existing low rate speech coding algorithms. In order to improve the LPC-10 vocoder, this algorithm uses 5 key technologies, namely multi-band mixed excitation, aperiodic pulse, adaptive spectral enhancement,pulse shaping filtering and residual harmonic spectrum processing; thus, the quality of the synthetic speech is improved.In this thesis, the basic knowledge and the principle of encoding and decoding of standard MELP algorithm are discussed in depth, and the corresponding formulas are deduced. On this basis, the 0.6Kb/s MELP algorithm is designed and implemented by structuring super frame and interpolation of intra-frame parameters. Moreover, synthetic speech waveforms,synthetic speech quality and algorithm complexity are adopted to evaluate this algorithm.Finally, a real-time speech communication system is realized based on MELP algorithm via C language on Linux system, and this system has been successfully transplanted into the Orange Pi PC, which is an ARM development board.