Using MATLAB for Analysis of TRAP system

TRAP, speech recognition, features extraction, estimated probability analysis

Using matlab for analysis of TRAP system. Especially, results from neural net classifiers. The utility of resulting figures is enhanced by the possibility of reading and displaying results from all critical bands at once. Resulting analyses are more focused on the reliability of classifiers than on the basic accuracy measure.

This article describes a Matlab function for reading and processing file outputs from a structure of classifiers. These classifiers - neural nets - are used in speech recognition based on temporal trajectories (TRAP) of energy in frequency bands. This nonstandard approach is introduced and the program is presented. The utility of resulting figures is enhanced by the possibility of reading and displaying results from all critical bands at once. Resulting analyses are more focused on the reliability of classifiers than on the basic accuracy measure. The first uses colors and their depth to display both cues, reliability and accuracy, in one informative picture. Others are focused on more precise measures, where it is possible to precisely define classifier mistakes.