14thAnnual Conference of the International Speech Communication Association

Lyon, France
August 25-29, 2013

Human Mouth State Detection Using Low Frequency Ultrasound

Farzaneh Ahmadi (1), Mousa Ahmadi (2), Ian Vince McLoughlin (3)

(1) Nanyang Technological University, Singapore
(2) University of Malaya, Malaysia
(3) USTC, China

for speech processing applications which is intrinsically robust to audible noise. Current solutions tend to rely on Doppler frequency shift to estimate movement of facial components during speech production. However, these approaches do not consider special attributes of LFUS propagation inside the vocal tract and consequently they do not discriminate between speech related and other forms of facial movements. This paper investigates the LFUS resonance characteristics of the vocal tract and uses these characteristics to develop a novel approach to determine the status of the mouth (open/partiallyopen/ closed). The proposed system provides high precision in detecting mouth status and provides a significant improvement to Doppler systems by being robust to variations of the direction of the face towards the ultrasonic source. The underlying paradigm is highly applicable in silent speech interfaces, multimodal voice activity detection, and hands-free control of rehabilitation devices for the disabled.

Full Paper

Bibliographic reference.  Ahmadi, Farzaneh / Ahmadi, Mousa / McLoughlin, Ian Vince (2013): "Human mouth state detection using low frequency ultrasound", In INTERSPEECH-2013, 1806-1810.