Audio Visual Speech Recognition (AVSR) presents important advantages in many applications when compared with Automatic Speech Recognition (ASR) standard approaches. Such is the case, for instance, when using some in-car devices or in Assistive Technology. AVSR also is a relatively recent topic in ASR, so its potential is far from being completely explored. Moreover, the AVSR approaches often require specially ingenious solutions to deal with the essential problem of combining efficiently the multiple feature streams that convey the linguistic information. Certainly these are some of the main arguments that have been contributing to the fact that AVSR has became my favourite interest in ASR.
Three main goals have been pursued in the AVSR topic, guiding the main activities that have been carried out: (1) keeping up with the state of the art, and in particular with proven approaches to essential problems; (2) applying the acquired (also through extensive experimentation based on efficient prototyping tools such as GMTK, HTK, CMU-LMtk and Matlab/Octave or libraries such as the OpenCV) know-how on the development of recognition modules and prototypes dedicated to concrete and useful applications; and (3) trying to create innovative solutions to solve or at least to mitigate some specific problems, eventually adapting and combining existing techniques or methods with new ones inspired on less related knowledge domains.
(last change: 31/10/2017)