8 kbits/s Speech coder with pitch adaptive vector quantizer

This paper proposes an 8 kbits/s speech coder with Pitch Adaptive Vector Quantizer (PAVQ), which efficiently quantizes the speech waveform in the time domain. The codebook of the PAVQ is classified into sub-groups each of which has the same maximum absolute magnitude (pitch peak) position. Each input vector is encoded by referring to the sub-group codebook having the same pitch peak position as that of the input vector, instead of referring to the whole codebook. If the codebook size of the PAVQ is equal to that of the conventional full-search VQ, computational complexity of the PAVQ is reduced much less than that of the conventional VQ. Computer simulation results show that the PAVQ is better in terms of speech quality than the conventional VQ under the condition of equal computational complexity. The PAVQ can be implemented in a single DSP and applicable memory IC chips.