Research On Low Bit Rate Speech Coding Based On Multi-band Excitation

Low bit rate speech coding is the main development tendency of speech coding now, due to the quick development and fierce competition of modern communication, how to utilize the channel effectively and economically must be important goal for study, low bit rate speech coding technology is an indispensable research direction in voice communication. Multi-band Excitation speech coding is a branch of low bit rate speech coding, it is also a hot research, there are a number of applications in the communication system.Beginning with MBE (Multi-band Excitation) speech model, compared with the conventional linear predictive speech coder, Multi-band Excitation speech encoder is more precise in the expression of excitation, it corresponds with the characteristics of actual voice better, the naturalness of the synthetic voice is higher. Multi-band Excitation speech coder transforms a short-time spectrum model of one frame speech signal into a product from a particular excitation spectrum and a system spectrum, according to the multiple of pitch frequency, excitation spectrum is divided into a number of harmonics, judging that each harmonic is unvoice or voice, excitation spectrums of unvoiced band are expressed with white noise spectrums, excitation spectrums of voiced band are expressed with periodic signals, various excitation spectrums of the band are added to get the entire excitation spectrum.Based on the INMARSAT-M IMBE vocoder model, the speech are analysed. INMARSAT-M IMBE vocoder is an improved algorithm based on MBE model, which is unique in pitch estimation, U/V judgment as well as the coding of the spectral envelope parameters. An improved pitch tracking algorithm based on MBE is proposed, which tracks only one frame data forward and backward respectively compared with the traditional one, aims to reduce calculations and upgrade the real-time performance of pitch detection. The result of simulation experiment shows that both the new algorithm and the traditional one can achieve good effect of pitch tracking, but the calculations and execution speed of the new algorithm are respectively reduced obviously and improved extremely. In the paper, some key steps in INMARSAT-M IMBE model analysis algorithm are described in detail, and attaching parts of the codes.Synthetize voice based on voice analysis. The speech synthesis is divided into unvoice ingredient synthesis and voice ingredients synthesis, overlying the two parts to get whole voice. Adaptive spectral enhancement is applied to MBE, the definition of synthetic speech is upgraded and the quality of synthetic speech is improved. The result of simulation experiment shows that the amplitudes of fundamental frequency and second harmonic of synthetic speech are both increased obviously by using adaptive spectral enhancement, the definition of synthetic speech is enhanced effectively.Using Matlab as the simulation platform, parameters are picked up from the original speech signals then speech synthetized, realizing the whole process, waveform diagrams of the original speech and the synthestic speech are drawed, then comparing the synthestic speech without adaptive spectral enhancement filter algorithm and the one with adaptive spectral enhancement filter algorithm, satisfactory results are obtained.