The problems associated with using multipulse excitations for low-bit-rate linear predictive coding (LPC) of speech are described, and a suggestion is made towards the realization of a 2.4 kb/s multipulse system. One of the reasons for the degradation of the speech quality of conventional LPC is that only one pulse in every pitch period is used as input to the synthesis filter. If more than one pulse per period is used, better speech quality can be realized. The method used for determining the pulse locations and the relevant amplitudes is the analysis-by-synthesis method. After determining the positions and amplitudes of all the pulses, the multipulse excitation signal is generated by filtering the spaces between the pulses by zeros. The synthesizer used with multipulse excitation is noted. The coding of the speech parameters is also discussed. The proposed system, working at 2.4 kb/s, is not yet implemented but a 4.8 kb/s system was simulated. The speech quality was satisfactory and there is room for a further reduction in the bit rate.
展开▼