EEG Based Voice Activity Detection

Abstract

Automatic speech recognition gain huge improvements in recent years. Deep neural networks used in speech recognition chain improved speech recognition accuracy to very high levels. Also these days, end-to-end speech recognizers are getting better. In contrast with these recent improvements, end user automatic speech recognition acceptance is quite low. It is due to fact that it does not work very well on long distances from microphone, yet. It is also impossible to use speech recognizers in places such as open offices or public places due to background noise. Another problem is that people do not want to disclose private or confidential information on loud. EEG based imagine speech recognizers could solve this acceptance rate. Overt speech recognizers may supplement speech recognizes from the microphone in situations where background noise is very high. Voice activity detector is a necessary component in Speech recognition chain and it is also true in EEG based speech recognition. Methods for voice activity detection from EEG signals are proposed in this paper.