The observation likelihood of silence: analysis and prospects for VAD applications

Abstract

This paper shows a research on the behaviour of the observa-tion likelihoods generated by the central state of asilenceHMM(Hidden Markov Model) trained for Automatic Speech Recog-nition (ASR) using cepstral mean and variance normalization(CMVN). We have seen that observation likelihood shows astable behaviour under different recording conditions, and thischaracteristic can be used to discriminate betweenspeechandsilenceframes. We present several experiments which provethat the mere use of a decision threshold produces robust re-sults for very different recording channels and noise conditions.The results have also been compared with those obtained by twostandard VAD systems, showing promising prospects. All in all,observation likelihood scores could be useful as the basis for thedevelopment of future VAD systems, with further research andanalysis to refine the results.