SRI is creating robust large vocabulary continuous speech recognition systems for the dual purposes of keyword detection and transcription of conversational speech. This project focuses on modeling conversational speech, combining DECIPHER hidden Markov modeling technology with neural network technology, creating tools for training, configuring, and rapid prototyping of wordspotting and recognition systems, and porting this technology to languages other than English.
Send comments and suggestions to webmaster web@speech.sri.com