VoxForge
Hello all,
I am trying to adapt the HMMs and i am in the phase of realigning the training data.
I am running the:
-->HVite -A -D -T 1 -l '*' -o SWT -b SENT-END -C
config -H macros -H hmmdefs -i adaptPhones.mlf -m -t 250.0 150.0
1000.0 -y lab -a -I adaptWords.mlf -S adapt.scp dict tiedlist
I've got to many warnings like this
-->''No tokens survived to final node of network''
This is not an Error but the data that are not surviving from the adapt.scp list when I'm running the HVite are too many (about a half of my data)!!
Does anyone know what is happening with all this "No tokes survived" and if this would be a problem for my adaptation?
Thanks in advance!!
--- (Edited on 7/23/2010 4:44 pm [GMT-0500] by tsinas) ---
>No tokens survived to final node of network
Listen to the audio file - it should give you the audio file name in the error message, and see if the transcription matches the audio file. If not, fix it or remove it from training.
If the transcripton matches the audio, then you may need to increase your -t parameters in HVite ( "-t 250.0 150.0 1000.0"). The HTK books explains how to do this. Adjusting these numbers is basically trial and error for a given corpus (at least that is what I do...)
>Error but the data that are not surviving from the adapt.scp list when I'm
>running the HVite are too many (about a half of my data)!!
Do all your audio files have the same sampling rate/bits per sample?
--- (Edited on 7/28/2010 4:04 pm [GMT-0400] by kmaclean) ---
Hi,
I am using HVite command
HVite -H hmm13/macros -H hmm13/hmmdefs -S test.scp -l '*' -i recout.mlf -w wordnet -p 0.0 -s 5.0 dict models1
The recout.mlf contains only #!MLF!#
I dunno what to do... Help!
--- (Edited on 6/15/2016 7:37 pm [GMT-0500] by ) ---