[SRILM User List] Computing nbest-error rate from HTK MLF files

[Nickolay V. Shmyrev]

On 24/04/2013 at 01:22 -0400, E wrote:
> Thanks for the response Andreas.
> 
> I will share my script once its ready. 
> 
> This "oracle" WER seems like a very crude way of computing nbest-error
> to me. Suppose a reference word is located in [0,  1] seconds, one can
> look at all the alternatives in the nbest list (all words that
> significantly overlap with reference word) and choose the word that
> best matches. 

You probably want to learn and use lattice oracle WER which finds the
best path in the lattice and can switch to different word variants on
the way at different times.

Overall n-best lists are not very good structure unless you are dealing
with long window rescoring with some advanced models like RNNLMs which
can't work on lattices. It's better to use lattices instead of n-best
lists wherever you can.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 198 bytes
Desc: This is a digitally signed message part
URL: <http://www.speech.sri.com/pipermail/srilm-user/attachments/20130424/866a9e30/attachment.bin>