I was quite surprised to see how low the WERs are for the new Common Voice corpus: https://github.com/kaldi-asr/kaldi/blob/master/egs/commonvoice/s5/RESULTS (4ish% TDNN)
Unfortunately, these result…
Dear all,
In our lab we are dealing with the scale of replacement matrices (i.e., empirical aa matrices) and we found that if we change the scale of a matrix (i.e., by multiplying all its elements by…