An improved CELP by the separate coding of pulsive and random residuals

Main drawback of the CELP coding is a S/N degradation when the duration of VQ of residuals becomes long in order to reduce bit-rate. The reason is an incomplete long time-prediction. Then, residuals are separated into pulsive and random components. After the separation, random residuals are coded by VQ with a random waveform code book, and pulsive components are coded in their timing and amplitude. The improved coding can transmit high quality of speech in less than 16 k bps.