Microsoft Research Announces Breakthrough Speech Recognition

Short and in Russian

A new technology, on which MR has been working for quite some time, was announced at Interspeech 2011, now taking place in Florence.
Using the hybrid context-sensitive deep neural network (CD-DNN-HMM) and George Dahl of the University of Toronto, the MR team managed to improve the quality and speed of speech recognition to " almost human "indicators. Due to the similarity of the structures of this neural network with 3D, the team achieved a significant increase in speed when using GPU computing.

[ Original news ]