Multimodal sensing for Subvocal speech recognition for Silent speech interfaces in future AR glasses

Inventor(s)

Melinda Anderson, Meta Platforms Technologies, LLCFollow
Thomas Lunner, Meta Platforms Technologies, LLCFollow
Ananta Narayanan Balaji, Meta Platforms Technologies, LLCFollow
Morteza Khaleghimeybodi, Meta Platforms Technologies, LLCFollow

Abstract

Subvocal speech occurs while someone reads a book or talks to oneself or murmurs. Unlike whispering where closer ones can hear speech, such voices cannot be heard by others nearby. Subvocal speech recognition is usually done with electromyograms (EMG) that are unique for each speaker. EMG electrodes are placed on the neck and facial muscles to record muscle activations while making subvocal speech. In addition, EEG electrodes can also be placed on the head to capture electrical signals generated by the brain when subvocal speech has been made. These electrical signals can then be interpreted using signal processing techniques for inferring speech.

Creative Commons License

This work is licensed under a Creative Commons Attribution 4.0 License.

Recommended Citation

Anderson, Melinda; Lunner, Thomas; Balaji, Ananta Narayanan; and Khaleghimeybodi, Morteza, "Multimodal sensing for Subvocal speech recognition for Silent speech interfaces in future AR glasses", Technical Disclosure Commons, (August 29, 2023)
https://www.tdcommons.org/dpubs_series/6205