Viseme Embeddings for Commonly Confused Words in Lip-Reading
Fenghour, S., Chen, D., Guo, K. and Xiao, P. (2023). Viseme Embeddings for Commonly Confused Words in Lip-Reading. the International Conference on Electrical, Computer, Communications and Mechatronics Engineering (ICECCME 2023). Tenerife, Canary Islands, Spain 19 - 20 Jul 2023 Institute of Electrical and Electronics Engineers (IEEE).
|Authors||Fenghour, S., Chen, D., Guo, K. and Xiao, P.|
Automating lip reading has attracted a lot of research attention in the last several years and there are a wide variety of applications where automated lip reading could be implemented, including forensic closed-circuit television footage or to assist those with language impairment. An obstacle for both professional lip-readers and automated lip-reading systems is not only the situation of different sounding words being confused because they produce identical lip movements, but also that word with similar lip movements can easily be confused because they have sequences of visemes that are similar. This paper proposes a library and corpus called Visemes2Vec for learning the embeddings of viseme sequences as a contribution and is to the best of our knowledge the first attempt at proposing library for visualising words that have similar lip movements. Visemes2Vec is analogous to Word2Vec which groups together that are semantically similar in meaning or Chars2Vec which groups together words that share common characters.
|Keywords||Lip-reading; Speech recognition; Viemes; Embedding; Neural networks; Principal component analysis|
|Journal||Proceedings of the International Conference on Electrical, Computer, Communications and Mechatronics Engineering|
|Publisher||Institute of Electrical and Electronics Engineers (IEEE)|
|Web address (URL)||http://www.iceccme.com/home|
|Accepted author manuscript|
File Access Level
|19 Jul 2023|
|Publication process dates|
|Accepted||20 Mar 2023|
|Deposited||20 Mar 2023|
Copyright © 2023 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.
Accepted author manuscript
1views this month
0downloads this month