The Impact of audible feedback on EMG-to-Speech Conversion

Research interest in speech interfaces that can function even when an audible acoustic signal is not present -- so-called \emph{Silent Speech Interfaces} -- has grown dramatically in recent years, as the field presents many barely exploded avenues for research and huge potential for applications in user interfaces and prosthetics.

EMG-to-Speech conversion is a type of silent speech interface, based on electromyography: It is the direct conversion of a facial electrical speech muscle activity signal to audible speech without an intermediate textual representation. Such a direct conversion approach is well suited to speech prosthesis and silent telephony applications and could be used as a pre-processing step to enable a user to use a regular acoustic speech interface silently. To enable these applications in practice, one requirement is that EMG-to-Speech conversion systems must be capable of producing output in real time and with low latency, and work on EMG signals recorded during silently produced speech.

The overall objective of this dissertation is to move EMG-to-Speech conversion further towards practical usability by building a real-time low-latency capable EMG-to-Speech conversion system and then use it to evaluate the effect of audible feedback, provided in real-time, on silent speech production.