Automatic Speech Recognition Using Kaldi

Speaker:
Ondřej Plátek
Abstract:
We present an extension of the Kaldi automatic speech recognition toolkit to support on-line speech recognition. The resulting recogniser supports acoustic models trained using state-of-the-art acoustic modelling techniques. (MFCC, MLLT+LDA, BMMI) As the recogniser produces word posterior lattices, it is particularly useful in statistical dialogue systems, which try to exploit uncertainty in the recogniser's output.  The evaluation is focused on on-line speech recognition in a dialogue system Alex for the public Czech transportation domain. The dialogue system is available at a public toll-free 800 899 998 line.
Length:
00:58:30
Date:
14/04/2014
views: 1979

Images:
Preview of img-017.jpg
Image img-017.jpg
Preview of img-025.jpg
Image img-025.jpg
Preview of img-034.jpg
Image img-034.jpg
Attachments: (video, slides, etc.)
54MB
1009 downloads
262MB
1980 downloads
56MB
992 downloads