Monday, April 14, 2014 -
13:30 to 15:00

Speech recognition using Kaldi

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.