VUS_Analysis_GUI25

This exercise utilizes four programs to train a Bayesian classifier and classify frames of signals.

This MATLAB exercise utilizes a set of four MATLAB programs to both train a Bayesian classifier (using a designated training set of 11 speech files embedded within a background of low level noise and miscellaneous acoustic effects (e.g. lip smack, pops, etc.)), and to classify frames of signal from independent test utterances as belonging to one of the three classes:
1. Class 1 – Silence/Background
2. Class 2 – Unvoiced Speech
3. Class 3 – Voiced Speech
using a Bayesian statistical framework as discussed in Section 10.4 of TADSP. The feature vector associated with each frame of signal consists of five short-time speech analysis parameters, namely:
1. short-time log energy,
2. short-time zero crossings per 10 msec interval,
3. normalized autocorrelation at unit sample delay,
4. first predictor coefficient of p = 12 pole LPC analysis,
5. normalized log prediction error of p = 12 LPC analysis.
The file '4.2 VUS Classification.pdf' provides a User's Guide for this exercise.