Speech classification embedded in adaptive codebook search for CELP coding

Former studies of coding methods based on speech classification usually utilize acoustic knowledge and techniques of speech recognition, or use a nonparametric approach without any apparent acoustic meaning in the classification. The authors propose a simple but effective method for speech classification which is based not only on the acoustic knowledge but also on the coding mechanism itself. A 3.6-kbit/s CELP (code-excited linear prediction) coder based on the classification method is designed. It can be shown that the speech classification coincides with the acoustic phenomena and is effective for CELP coding with adaptive codebook search. Simulation results show that the quality gets better as the computational complexity is reduced, and the pitch contour of the synthesized speech is smoother than that produced by conventional CELP coders.<>