An open source speech synthesis module for a visual-speech recognition system

A Silent Speech Interface (SSI) is a voice replacement technology that permits speech communication without vocalisation. The visual-speech recognition engine of the proposed SSI is based on vocal tract imaging. The system aims to give the laryngectomised speaker the opportunity to speak with his/her original voice. This paper presents the speech synthesis module of a SSI that uses the open-source MaryTTS (Text-To-Speech). The visual-speech recognition engine of the SSI outputs a text sentence, which is imported to the speech synthesis module in order to synthesise speech in French or English. A new module of phonetic transcription has been developed and integrated into MaryTTS. In addition, English and French semi-HMM (Hidden Markov Models) model voices have been built. The SSI can be remotely controlled using a mobile device and the new voices are installed in a Web Server.