Assignment 4, Weeks of March 22 and 29: Report Due: Tues. April 6
- Reading: chapters 8 and 9 of the text
For a brief history of speech synthesis,
click here.
- Problems and Lab work:
- Work Problems 8.10 (number of function calls in recursive
implementation FFT) and Problem 9.4 (find a z-transform).
- Demonstrate the "precedence effect": wherein the brain
assigns a direction to a sound source by comparing
arrival times. Try periodic clicks in earphones with
relative delay between channels. If you have time,
study the effects of the recurrence frequency of the
clicks, loudness, and any other parameters you can think of.
- The following are commonly accepted formant frequency locations
for male vowel speech sounds: [click
here for formant table].
Look at the speech spectrographs of some actual speech sounds
and see how these values compare. Try to synthesize some of these
vowels by sending a pulse train (or buzz if you have one) into a
cascade of three resons with the corresponding peak frequencies.
Hint: move the pitch during a test. Question: why might this help?
Extra credit: Try to make the vowels obtained this way more realistic
by adding a filter in cascade that shapes the pulses so that they
look like "glottal pulses". Add another cascade filter that models
impedance matching from the vocal tract to free wave radiation.
Look these up under "source-filter model" in a book on digital
processing of speech synthesis, like the following for example:
- F. J. Owens, Signal Processing of Speech, Macmillan, 1993.
- L. R. Rabiner and Schafer, R.W., Digital Processing of
Speech Signals, Prentice Hall, 1978.
- L. R. Rabiner and B.-H. Juang, 1993, Fundamentals of Speech
Recognition, Prentice Hall, 1993.
- S. Rosen and Howell, P., Signals and Systems for Speech and
Hearing, Academic Press, 1991.
- Experiment with Linear Predictive Coding.