Improved speech analysis for glottal excited linear predictive speech coding
Adewale Olatunji · 2004
We describe a new method for extracting the voice source and vocal tract impulse response from voiced speech for glottal excited linear predictive speech coding. The method utilises a mixed phase model of speech and dynamic gain control to deconvolve speech signals into source and filter components. Voiced speech segments are analysed for pitch synchronously and coded with a glottal codebook of 32 entries, while the processing of unvoiced speech and silence intervals is done with a stochastic codebook. The method also exploits a high correlation among glottal pulses to simplify the algorithm computational complexity and to improve codebook search time. Results of coding both male and female speech with the new method are presented.