Speaker independent continuous speech recognition

Abstract

This report presents a detail study on psychoacoustic modeling for feature extraction for robust continuous speech recognition. In an automatic speech recognition (ASR) system, feature extraction is critical to determining the recognizer’s performance. The most popular feature vectors for ASR are Mel Frequency Cepstral Coefficients (MFCC). However, it is also well known that its performance drops dramatically under noisy condition. One of the objectives of this research is to improve on the robustness of a continuous speech recognizer.