[MUD-Dev] [TECH] Voice in MO* - Phoneme Decomposition and Rec onstruction

ceo at grexengine.com ceo at grexengine.com
Sun Jun 2 08:19:59 New Zealand Standard Time 2002

```On Wed, May 29, 2002 at 11:55:31AM -0500, Robert Zubek wrote:
> From: Eli Stevens [mailto:listsub at wickedgrey.com]

>> What I would like to know is if there is a good way (i.e. not
>> brute force) to figure out when a wave has finished and is now
>> repeating.

> Look into spectral analysis techniques, such as the Fourier
> Transform. It will be able to compute a frequency distribution for
> any signal sample, which you can then analyze.

> Also, a good overview of current research in speech understanding
> is:

>   "Survey of the State of the Art in Human Language Technologies"
>   http://cslu.cse.ogi.edu/HLTsurvey/

Another source you might find helpful:

Information Theory And Coding
- http://www.cl.cam.ac.uk/Teaching/2001/InfoTheory/

At bottom of page there is a postscript file of complete lecture
notes.  First half of course is all about Entropy and Information
Content, and how they are the inverse of each other...well
written, easy to understand, and sort of required for
understanding the second half but...

The second half is IMHO incomprehensible; others ahve found it
OK, if hard going. Have a look at where it starts talking about
Fourier onwards, and decided if its worth looking at the first
half!

Section titles:

Overview and historical origins: foundations and uncertainty.

Mathematical foundations; probability rules; Bayes' theorem.

Entropies defined, and why they are measures of information.

Source coding theorem; prefix, variable-, and fixed-length codes.

Channel types, properties, noise, and channel capacity.

Continuous information; density; noisy channel coding theorem.

Fourier series, convergence, orthogonal representation.

Useful Fourier theorems; transform pairs. Sampling; aliasing.

Discrete Fourier transform. Fast Fourier Transform Algorithms.

The quantised degrees-of-freedom in a continuous signal.

Gabor-Heisenberg-Weyl uncertainty relation. Optimal ``Logons''.

Kolmogorov complexity and minimal description length