Re:Researchers achieve speech recognition milestone
by WilliamDhalgren on Thursday September 15, 2016 @12:05AM
(#12503333)
So this model won't be free software??? Odd, and bummer...
Also I'm note sure an error reduction of 20% (1-6.3/7.8) is to be considered small; depends on the particular challenge really. Like, sentiment analysis only starts to get interesting above 80% on some dataset, as much can be guessed correctly in very naive ways..
Human lvl on this task is estimated to be ~4% so we have quite a lot of ground to cover still..
Re:Researchers achieve speech recognition milestone
by timgws on Thursday September 15, 2016 @01:44AM
(#12503730)
There are 33% less errors with the Microsoft solution then with Kaldi... one could say that is quite significant.
Re:Researchers achieve speech recognition milestone
by afsina on Monday September 19, 2016 @09:59AM
(#12531483)
Relative decrease in WER is not so significant for lower percentages. How about "we make 6 errors on 100 words but Kaldi makes 7".
Re:Researchers achieve speech recognition milestone
by dave168 on Friday September 16, 2016 @01:56AM
(#12512130)
It is cool anyone can use CNTK to produce something similar now
Re:Researchers achieve speech recognition milestone
by nshm on Thursday September 15, 2016 @03:05AM
(#12504003)
23% only
Re:Researchers achieve speech recognition milestone
by visarga on Thursday September 15, 2016 @12:49AM
(#12503502)
Maybe they can distillate the ensemble in a compact and efficient version for production.
Re:Researchers achieve speech recognition milestone
by danielmorozoff on Wednesday September 14, 2016 @07:44PM
(#12502209)
Could you please provide a link?
Re:Researchers achieve speech recognition milestone
by nshm on Wednesday September 14, 2016 @07:52PM
(#12502263)
https://github.com/kaldi-asr/kaldi/blob/master/egs/fisher_sw...
In Microsoft paper http://arxiv.org/pdf/1609.03528v1.pdf Table 5 it's a line "Povey et al. [19] LSTM". http://www.isca-speech.org/archive/Interspeech_2016/pdfs/059... Interpolate it to RNNLM column and you'll get 7.8. See also http://www.isca-speech.org/archive/Interspeech_2016/pdfs/047...
Also, major issue with this kind of research is that they combined several systems in order to get best results. Most practical systems don't use combinations, they are too slow.