Research On Very Low Bit Rate Speech Coding Algorithm

Speech Coding is of great importance in digital communication systems. At the situation where the transmission rate is limited strictly, Very Low Bit Rate Speech Coding (LBRSC) is especially significant.As an important algorithm of LBRSC, the Mixed Excitation Linear Prediction (MELP) algorithm which was choosen as U.S. Federal Standard has got quite good speech quality at the rate of 2.4Kb/s, but there are still some perceivable problems, particularly around non-stationary speech segments and in the aspect of coding efficiency.In this thesis, MELP algorithm is deeply studied. In order to higher the coding efficiency, the interpolation algorithm of invariable frame rate is presented; Based on the VAD algorithm in G.729B, the BD-VAD algorithm is promoted; After the investigation and analysis of the inter-frame parameter correlation, an interpolation algorithm based on the spectral slope constraint is presented to lower the coding rate, meanwhile, the speech quality and the operating complexity is similar to MELP algorithm.The transmission bit rate of this speech coding/decoding system is lowered to 300-800 bps. After comparing and subjectively evaluating the reconstructed speech, it is concluded that the performance of this system approaches to that of 2.4Kb/s MELP algorithm which is in the Federal Telecommunication Recommendation.