Research On Low Bit Rate Speech Coding

Mixed Excitation Linear Prediction's development is based on the dual excitation LPC vocoder. There are 5 new technologies which are multi-band mixed excitation, aperiodic pulses, adaptive spectral enhancement, pulse dispersion filtering and residual harmonic treatment, so it can make the synthesized voice more naturally, at the same time the bit rate and the bit error are becoming lower. With so many good features, MELP has a good prospect in the field of mobile communications and satellite communications.MELP algorithm is divided into the coding part and the decoding part. According to the short-term stability of the voice, coding part extract various feature parameters from the speech signal in period time, and then quantify the coding of these parameters, finally the important parameters are protected through using error correction coding and all of the parameters are transferred into the signal channel. Decoding section includes decoding the received parameters, calculating periodic pitch, generating mixed excitation signal, doing adaptive spectrum enhancement, linear prediction synthesis, gain adjustment and pulse dispersion filtering to mixed excitation signal and generating the speech signal.In the current communication systems, real-time speech coding is usually achieved through the digital signal processor (DSP). TI has launched a highly integrated, high performance and low power fixed-point DSP processor named TMS320VC5509A, its cpu can run on the highest frequency of 200MHz. This paper uses TMS320VC5509A as its experimental platform and TLV320AIC23 as its stere audio codec. The hardware platform is used to capture, playback speech signal and run MELP algorithm.Firstly, the research briefly overviewes the significance and the background of the subject, and then deliberates the principle of MELP algorithm. Then this paper introduces 2.4kbps MELP's transplantation and its implementation in the fixed-point DSP chip TMS320VC5509A. As to the hardware, it describes register configuration of McBSP, DMA, TLV320AI23, and depictes the connections and communications among them. As to the software, it introduces the transplantation and optimization of MELP algorithm. Finally it compares the results between the previous and the optimized encoding, the result verifies the practical feasibility of the design.