This project implements a
speech synthesizer utilizing a source-filter model that resembles actual
human speech production. Literature survey on the anatomy of human speech
production, source-filter models, and all-pole filters of linear prediction
coefficients are performed. While we discovered many ways to generate
synthetic speech, we chose an all-pole synthesizer derived from linear
predictive coding analysis(LPC). This method is chosen because linear
prediction enables us to characterize the vocal tract very efficiently.
(Also, we just wanted to learn about something we've never seen before.)
With some coding in MATLAB, we demonstrated that we could make our computer
produce the sounds /s/, /a/, and /i/. The concatenation of these sounds
could produce utterances such as "saucy" and "seesaw."
Although the words are far from natural sounding speech, we have demonstrated
that very easy techniques are powerful enough to do some very simple
speech synthesizing.