dc.contributor.author | Mwema, Wilfred N | |
dc.date.accessioned | 2013-05-04T08:25:44Z | |
dc.date.available | 2013-05-04T08:25:44Z | |
dc.date.issued | 1994 | |
dc.identifier.citation | Masters thesis University of Nairobi 1994 | en |
dc.identifier.uri | http://erepository.uonbi.ac.ke:8080/xmlui/handle/123456789/18913 | |
dc.description.abstract | The multi-pulse excitation Linear predictive Coding
lMPE-LPC) system produces synthetic speech that is more
natural sounding than the classical linear predictive coder.
In the the MPE-LPC system, the excitation signal is modeled
with a few pulses per frame of speech regardless of whether
the frame IS voiced or unvoiced. ~he quality of the
synthesized speech improves with the number of pulses used
per frame. Pulses are computed by minimizing the weighted
square error between the original speech and the synthetic
speech. Conceptually, this is a simple problem. However,
locating the pulses in the excitation signal can be quite
involving when the number of pulses is large.
In this study the MPE-LPC system is described and
simulated on an IBM PC30XT computer. A computation reduction
method in the pulse search process is proposed and compared
with other existing techniques. Results show that this
method compares well with earlier methods.
The effect of the MPE-LPC system parameters on the
quality of the reconstructed speech has been investigated and
explained. Pulse density reduction by pitch prediction has
also been attempted using a one-tap predictor. Results
obtained show that pitch prediction improves the quality of
the reconstructed speech but not significantly. | en |
dc.language.iso | en | en |
dc.publisher | University of Nairobi | en |
dc.title | Multi-pulse excitation linear predictive coding of speech signals | en |
dc.type | Thesis | en |
local.publisher | Department of Electrical and Electronic Engineering | en |