A Novel Two-Stage System for Spotting Fricative and Plosive Regions from Continuous Speech

摘要

In this paper, a novel method for spotting the fricative and plosive regions from continuous speech is discussed. A two-stage recognition system is designed for spotting and verifying the fricative and plosive region. In the first stage, a DNN based broad classifier is used to convert the input speech signal to corresponding broad phoneme classes. Thus fricative and plosive regions are shortlisted. In the second stage, verification of these regions are done using different signal based features. Spectral centroid is used to verify the fricative regions and difference in spectral spread is used to verify the plosive regions. The proposed idea is useful for spotting fricative and plosive regions in isolated words as well as in continuous speech. Studies are done using Malayalam read speech data and with TIMIT database.