Posts: 4
Threads: 4
Joined: Feb 2010
Posts: 261
Threads: 2
Joined: Feb 2010
This system uses modified diphones to train on the regions of greatest spectral stability as well as spectral change. In contrast, the previous methods segment the data into monophones, biphones, or triphones and train on each sub-phone unit in several broad-category contexts. regions of spectral stability are accounted for but their presence in the word models is not absolutely necessary. Empirical evidence for the advantage of this new method is seen by the 13 % reduction in word-level error. This was in the test set by OGI Numbers corpus. Comparitive study was conducted on a baseline system that used context-independent monophones and context-dependent biphones and triphones.
report:
[attachment=1916]
Posts: 6,566
Threads: 1,107
Joined: Jul 2011
Posts: 6,566
Threads: 1,107
Joined: Jul 2011
Posts: 6,566
Threads: 1,107
Joined: Jul 2011