A voice activity detector based on cepstral analysis
Jack Haigh, John S. D. Mason · 1993
This paper proposes a new approach to speech end-point detection based on cepstral analysis. The algorithm is based on explicit (static) modelling of speech and non-speech, and decisions are made on each incoming (overlapped) cepstral frame, according to model similarity scores. The cepstral analysis provides excellent levelindependence, meaning that parameter adjustment, decision thresholds etc, are unnecessary. A high degree of robustness to additive noise is demonstrated, even though the models are static. Accurate end-points are recovered with SNR levels of 0dB. Keywords: speech analysis, end-point detection, voice activity detection, robustness. 1 INTRODUCTION Many early voice-activity-detection (VAD) algorithms were based on a combination of shortterm energy and zero-crossing-rate (ZCR) measurements, [1], [2] and [3]. Such algorithms detect speech by relying on the fact that an increase in energy is likely to occur somewhere between the two ends of a word or `talkspurt'. Havi...