Investigation on vector quantized multipulse excitation with fixed pattern of positions

Abstract

The authors consider speech coding based on the vector quantization of the multipulse excitation with a fixed pattern of positions (MPE-FPP). Since in this type of excitation signal the positions of the pulses are known, a codebook can be easily obtained by applying the vector quantization design technique directly to a training sequence of the multipulse excitation generated by a conventional multipulse linear prediction (MPE-LPC) speech coder with restrictions on the pulse positions. Informal listening tests have shown that the vector quantized multipulse excitation with fixed pattern of positions (VQMPE-FPP) is very efficient in producing natural-sounding speech at a bit rate as low as 6.7-7.1 kbit/s.<>