An LPC Excitation Model using Wavelets

Langi, Armein Z.R. ( Research Center on Information and Communication Technology Information Technology RG, School of Electrical Engineering and Informatics Institut Teknologi Bandung, Jalan Ganeca 10, Bandung, 40116, Indonesia)

Journal of Engineering and Technological Sciences
Vol 40, No 2 (2008)

This  paper  presents  a  new  model  of  linear  predictive  coding  (LPC) excitation  using  wavelets  for  speech  signals.   The  LPC  excitation   becomes  a linear combination of a set of self-  similar, orthonormal, band-pass signals with time localization and constant bandwidth in a logarithmic scale. Thus, the set of the  coefficients  in  the  linear  combination  represents  the  LPC  excitation.  The discrete  wavelet  transform  (DWT)  obtains  the  coefficients,  having  several asymmetrical  and  non-uniform  distribution  properties  that  are  attractive  for speech processing and compression. The properties include magnitude dependent sensitivity, scale dependent sensitivity, and limited frame length, which can be used  for  having  low  bit-rate  speech.  We  show  that  eliminating  8.97%  highest magnitude  coefficients  degrades  speech  quality  down  to  1.49dB  SNR,  while eliminating  27.51%  lowest  magnitude  coefficient  maintain  speech  quality  at  a level of 27.42 dB SNR. Furthermore eliminating 6.25% coefficients located at a scale associated with 175-630 Hz band severely degrades speech quality down to 4.20 dB SNR. Finally, our results show that optimal frame length for telephony applications is among 32, 64, or 128 samples.