Isolate speech recognition based on time-frequency analysis methods

Abstract

A feature extraction method for isolate speech recognition is proposed, which is based on a time frequency analysis using a critical band concept similar to that performed in the inner ear model; which emulates the inner ear behavior by performing signal decomposition, similar to carried out by the basilar membrane. Evaluation results show that the proposed method performs better than other previously proposed feature extraction methods when it is used to characterize normal as well as esophageal speech signal. © 2009 Springer-Verlag Berlin Heidelberg.