Abstract:

This paper presents a model of the Multimode Multi-Band Excitation (MBE) joined with the Linear Prediction vocoder. Compared with MBE coding, this vocoder has some improvements in two ways. First, the Linear Predictive Mode is combined with Spectrum Estimation, and then the linear predictive coefficients are denoted and quantized by Line Spectral Frequency parameters (LSF), which solves the problem of variable dimensions in spectrum estimation in the MBE system. Second, in Unvoiced/Voiced (U/V) Decision, the number of sub-bands in one frame is fixed, ahd then the decisions are divided into different modes. LSF vectors are quantized by different codebooks to improve the quantization result, according to the different statistic distributions of the LSF vector in different U/V determinative modes. Simulation by the C programming language demonstrates that the synthesized speech Intelligibility is high and that the speech waveforms in the time domain and spectrograms both fit well.

Key words: vocoder, multimode multi-band excitation, linear prediction, line spectral frequency