Speech Activity Detection from EEG using a feed-forward neural network

Brain-computer interface (BCI) technologies connect brain sensing devices to computers where these signals are analyzed and processed. This makes the connection between the central nervous system and the computer. The most widely used BCI technologies include electroencephalograph. Electroencephalography brings the possibility of a mobile BCI system usable in life. However, the creation of such a mass market system still requires solving many fundamental problems. Automatic speech recognition gains 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 the fact that it does not work very well on long distances from a 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. Electroencephalography based imagine speech recognizers could solve this acceptance rate. Speech recognizers may supplement speech recognizes from the microphone in situations where background noise is very high. Speech activity detector is a necessary component in the Speech recognition chain and it is also true in EEG based speech recognition. Methods for speech activity detection from EEG signals are proposed in this paper.