Pitch synchronous split-band LPC (PS-SBLPC) vocoder

A pitch-synchronous split band LPC (PS-SBLPC) speech coder is proposed. In this new paradigm, harmonic analysis is carried out on individual pitch cycle waveforms (PCWs) rather than using a large window. PCWs are identified using a trapezoidal window search performed on a modified time envelope signal. In order to achieve a fixed rate coder the PCW parameters are jointly quantised using a combined interpolation and quantisation routine. Combining interpolation and quantisation allows for high correlation between successive PCWs to be exploited, without subjecting rapid transitions to time smoothing. During speech synthesis, no interpolation is applied, as parameter smoothing is provided during the quantisation. Simulation results comparing the PS-SBLPC model with the SB-LPC model show that the quality of the PS-SBLPC speech signal is significantly better than that of the split band LPC (SB-LPC). Initial results have shown that quantisation optimisation leads to vast improvements in speech quality during speech transitions.