Continuous optical automatic speech recognition by lipreading

Abstract
We describe a continuous optical automatic speechrecognizer (OASR) that uses optical information fromthe oral-cavity shadow of a speaker. The systemachieves a 25.3 percent recognition on sentences havinga perplexity of 150 without using any syntactic,semantic, acoustic, or contextual guides. We introduce13, mostly dynamic, oral-cavity features used foroptical recognition, present phones that appear opticallysimilar (visemes) for our speaker, and presentthe recognition results for our...

This publication has 9 references indexed in Scilit: