A novel wireless tongue tracking system for speech applications

Algorithm;Articulograph;Biomedical device;Deep learning;Machine learning;Magnetometers;Motion tracking;Neural network;Nonlinear optimization;Permanent magnet localization;Speech;Tongue tracking;Wearable

Sebkhi, Nordine ; Inan, Omer Electrical and Computer Engineering Anderson, David Santus, Nina Bhatti, Pamela Klein, Benjamin ; Inan, Omer

University:Georgia Institute of Technology

Department:Electrical and Computer Engineering

关键词: Algorithm;    Articulograph;    Biomedical device;    Deep learning;    Machine learning;    Magnetometers;    Motion tracking;    Neural network;    Nonlinear optimization;    Permanent magnet localization;    Speech;    Tongue tracking;    Wearable;   

Others  :  https://smartech.gatech.edu/bitstream/1853/62309/1/SEBKHI-DISSERTATION-2019.pdf

美国|英语

来源: SMARTech Repository

【 摘 要 】

The technological innovation of MagTrack is its ability to track the position of a magnetic tracer wirelessly, and with an accuracy of less than 2 mm, by implementing a novel permanent magnet localization method. Combining its wireless and high-accuracy tracking capability with the millimetric dimensions of the tracer, MagTrack is very well-suited for tongue tracking. Compared to commercially available tongue tracking systems, MagTrack has many of their advantages but without most of their shortcomings. For instance, MagTrack can be embedded in a headset to be portable, is affordable by using off-the-shelf components and custom-designed parts that are 3D-printed in our lab, and the tongue can be tracked wirelessly, thus enabling it to move without any hindrance.Having access to tongue motion is important for many applications. Chief among them is the treatment of speech sound disorders in which the errant placement of the tongue is responsible for the decreased speech intelligibility. A preliminary study was conducted to assess the feasibility of MagTrack to be used in gamified visual feedback that would help patients better practice correct placement of their tongue by providing objective and quantitative measures of speech performance based on the error between their tongue placement and carefully-selected visual targets. Another promising application for MagTrack is as the input of a silent speech interface that can recognize speech solely from tongue motion to generate synthesized voice for people that cannot produce sounds. Pilot human studies were conducted to obtain a preliminary assessment of speech recognition accuracy and the quality of synthesized speech from tongue motion recorded on hundreds of utterances by MagTrack. Finally, since MagTrack is an improved version of the Tongue Drive System, it has the potential to be a practical assistive technology to help people with quadriplegia to regain autonomy for some tasks of their daily life such as driving their powered wheelchair, or even control their computer and mobile devices without the help of their caregivers.