VoxForge
Hi,
My dictionary have only ten digits (oh, zero, one..., nine) from TIDIGITS database, I don't know if I can use the fulllist in your web site. I tried to use it and I received this error when I run this command in HTK book:
HHEd -B -H hmm12/macros -H hmm12/hmmdefs -M hmm13 tree.hed triphones1 > log
ERROR [+2662] FindProtoModel: no proto for y-uh+r in hSet
I don't know where is this error comes from and how to fix it, can you help me?
Thanks a lot
Thanh
Hi Thanh,
>ERROR [+2662] FindProtoModel: no proto for y-uh+r in hSet
This error is saying that the HTK training process found a word in [your grammar file -or- your pronunciation dictionary] with a triphone that it cannot find in the training set (i.e. prompt files and speech audio files).
Note: Step 1 through 8 in the tutorial creates a monophone acoustic model using the phones in your grammar file. Steps 9 &10 create triphones from the *entire* pronunciation dictionary (not just the words in your grammar file)
If, as you say, your dictionary only has ten digits, then it is likely that you have a word in your grammar (it might be "your") that has a triphone that is not included in the training set.
>My dictionary have only ten digits (oh, zero, one..., nine) from TIDIGITS database, I don't know if I can use the fulllist in your web site.
likely not - the fulllist file with the VoxForge tutorial uses pronunciations based on a modified version of the pronunciation dictionary included with ISIP Switchboard corpus.
You should be able to create your own based on the information in Step 10 in the VoxForge tutorial.
Hope that helps,
Ken