V M Thakare, Voice Activity Detector and Noise Trackers for Speech Recognition System

The well known fact is that the performance of the Speech Recognition System degrades drastically in Adverse Environments. Additive noise is one of the major element of adverse noisy environment. Detecting voiced, un-voiced or silent speech signal in noisy environment is not an easy task. A voice activity detector (VAD) is effective when the noise is stationary; it often fails when the noise statistics change during speech presence. Moreover, accurate voice activity detection under very low signal-to-noise-ratio (SNR) conditions is not trivial. The noise estimate can have a major impact on the quality of the enhanced signal. If the noise estimate is too low, annoying residual noise will be audible, while if the noise estimate is too high, speech will be distorted resulting possibly in intelligibility loss. The different VAD methods noise tracking approaches need to implement to enhance the speech signals collected through microphone in human-computer interaction. Although such approaches might work satisfactorily in stationary noise (e.g., white noise), but they may not work well in more realistic environments (e.g., in a restaurant) where the spectral characteristics of the noise might be changing constantly. Hence there is a need to update the noise spectrum continuously over time and this can be done using noise-tracking algorithms. This paper presents some Voice Activity Detecting (VAD) and noise tracking approaches that will help to improve the performance of speech recognition system in adverse environment for human-computer interaction