D306 SottoVoce: A Silent Speech Interaction System using Deep Neural Networks

Rekimoto Laboratory, The University of Tokyo / Sony Computer Science Laboratories, Inc.

The availability of digital devices operated by voice is expanding rapidly. However, the voice interface has problems such as difficulty in use in public places and reduced accuracy due to noise. SottoVoce is a system that detects the silent voice of a user using ultrasonic echo images. The user's utterance content is recognized by deep learning from intraoral information observed by an ultrasound imaging probe attached to the underside of the chin.

Technical Specification

The availability of digital devices operated by voice is expanding rapidly. However, the voice interface has problems such as difficulty in use in public places and reduced accuracy due to noise. SottoVoce is a system that detects the silent voice of a user using ultrasonic echo images. The user's utterance content is recognized by deep learning from intraoral information observed by an ultrasound imaging probe attached to the underside of the chin. We are researching this project with the aim of developing it as an input method for future wearable computers and as a support technology for people who have difficulty with voiced speech.