A Voice Activity Detector Based on Noise Spectrum Adaptation and Discrimination Information for Automatic Speech Recognition System

In this paper, an adaptive voice activity detector (VAD) is proposed, which is successfully implemented in a MFCC based speech recognition system. The proposed VAD describes a novel scheme of detecting speech presence/absence by tracking the higher portion of speech power spectrum and judging the discrimination information. The VAD will adjust judgment threshold adaptively. An automatic speech recognition (ASR) system is designed and conducted. In the ASR system the input feature is converted to a sub-word unit index sequence using the Bhattacharyya distance. Evaluation tests are carried out using noise database NOISEX-92 and speech database YOHO Corpus. Experimental results show that the proposed algorithm manages to achieve very impressive improvements.