MBR does not work for the phonetisaurus-0.7.5

[GoogleCodeExporter]

What steps will reproduce the problem?
1) Compile OpenFST and phonetisaurus
2) Use the toy files from fsmnlp-tutorial.tgz
3) Train the model using the line-by-line guide from 
http://code.google.com/p/phonetisaurus/wiki/FSMNLPTutorial
4) Try to use the MBR decoding

What is the expected output? What do you see instead?
I see "Segmentation fault", instead of output equivalent to the tutorial

What version of the product are you using? On what operating system?
phonetisaurus-0.7.5 on linux, Debian 6.0.6, gcc (Debian 4.4.5-8) 4.4.5

The source code from the tutorial provides the expected results.

Original issue reported on code.google.com by jtr...@gmail.com on 10 Jan 2013 at 10:52

[GoogleCodeExporter]

Will check it out and update.  In the meantime, keep in mind it is actually 
pretty slow.  The tradeoff in terms of speed vs. accuracy is probably not worth 
it.  

Original comment by Josef.Ro...@gmail.com on 21 Jan 2013 at 2:28

• Changed state: Accepted

[GoogleCodeExporter]

[deleted comment]

[GoogleCodeExporter]

the omega version should function, but i still think it is out of line with the 
tutorial version.  im a bit preoccupied at the moment so it will be at least 
another week before i have time to check it out in detail.  in the meantime id 
stick with the fsmnlp tutorial distro if you really need it.

Original comment by Josef.Ro...@gmail.com on 22 Jan 2013 at 1:14

• Changed state: Started

[GoogleCodeExporter]

I'm closing this as it should work correctly, just very slowly, and the 
trade-off does not seem justified given that training a mixture of a couple of 
models (KN+ME+RNN) takes only an hour or so for 1M+ entries and seems to yield 
as good or better results in most cases. 

Original comment by Josef.Ro...@gmail.com on 22 Apr 2014 at 8:28

• Changed state: Done