|
From: Josef N. <jos...@gm...> - 2014-01-07 11:20:41
|
Hi, (duplicate since I did not reply to the list the first time) This tool was not designed for and is almost certainly not compatible Kaldi. Kaldi employs some additional symbol management for the backoffs, which are not supported in my toolchain (I made it before Kaldi existed as a public project). You will have similar problems if you try to use the OpenGRM ngram tools without modification (but the google tool does the best hole-plugging). In general, default optimization commands (det, min, etc.) do not guarantee anything regarding symbol synchronization. I believe Kaldi does some extra work in this regard. Using positional phonemes can make it easier to retrieve the boundaries (I think there are some tools in Kaldi to do this after the fact). Probably it is not appropriate for this list since it is not compatible with Kaldi. -Joe On Tue, Jan 7, 2014 at 10:22 AM, E <oth...@ao...> wrote: > Hello, > > This is probably off-topic but I hope someone here knows answer. I am > converting ARPA LM to FST using "transducersaurus". > The command that I use is C*min(det(L*(G))) > > My problem is, the *.fst.txt file that is created has output symbols at > unexpected locations. > > I expect the output symbols to be <eps> at all phonemes in the word except > last phoneme and output symbol = "word" for only the last phoneme. > > But what I'm seeing is that sometimes output symbol = "word" even for > phonemes that are not last phonemes in the word. > > I wonder what the issue is and how to properly identify word end boundary in > an FST. > > Let me know if sharing my files will help in diagnosis. > > Thanks, > Ethan > > ------------------------------------------------------------------------------ > Rapidly troubleshoot problems before they affect your business. Most IT > organizations don't have a clear picture of how application performance > affects their revenue. With AppDynamics, you get 100% visibility into your > Java,.NET, & PHP application. Start your 15-day FREE TRIAL of AppDynamics > Pro! > http://pubads.g.doubleclick.net/gampad/clk?id=84349831&iu=/4140/ostg.clktrk > _______________________________________________ > Kaldi-users mailing list > Kal...@li... > https://lists.sourceforge.net/lists/listinfo/kaldi-users > |