VoxForge
Hi all,
I have to generate the phoneme (or word) confusion matrix by HResults of HTK. I have tried but it wasn't work. Does anyone has experience of doing this?
Thanks for your help.
Do Cong Thanh
--- (Edited on 7/18/2008 6:13 am [GMT-0500] by docongthanh) ---
Hi Do Cong Thanh,
The HTK Book says:
HResults contains a number of other options. ... a confusion matrix can be generated by setting the -p option
What errors were you getting? Please re-run the HResults command with the "-A -D -V -T 1" parameters (in addition to the ones you used when you got the error), and post the console ouput.
thanks,
Ken
--- (Edited on 7/23/2008 12:03 am [GMT-0400] by kmaclean) ---
Hi Ken,
I have used the -p option like you have said but I didn't used the "-A -D -V -T 1" parameters. I'm not at my office at the moment. I'll come back from holiday next week so I will retry and we will discuss more about this.
Thanks.
Thanh
--- (Edited on 7/28/2008 6:52 am [GMT-0500] by docongthanh) ---
In fact, even when I use the -A -D -V -T 1 and the -p options, there are nothing new about the results. I used the command like following:
HResults -I -p words_test.mlf monophones1 recout.mlf (I use the monophones HMMs for recognition)
words_test.mlf contains word level transcriptions. So when I add the -p option which is the option for generating phoneme confusion matrix, there will be the error message like "Label SIL not in list [0 22]". So I think the problem come from this transcription file. Maybe I have to use a phone level transcription file here. I don't know if we want to generate phoneme confusion matrix, we have to use phone level transcription or not? Further, recount.mlf was generated by HVite and it is in the form of word level, I don't know if we can generate recount.mlf in the phone level transcription?
Regards,
Thanh
--- (Edited on 8/6/2008 10:24 am [GMT-0500] by docongthanh) ---
Hi docongthanh,
>even when I use the -A -D -V -T 1 and the -p options, there are nothing new about the results
I wanted you to use the -A -D -V -T 1 options and post the console output here so I could know which version of HTK you are using...
>I don't know if we want to generate phoneme confusion matrix, we have to
>use phone level transcription or not?
Don't know off-hand... give it a try and let us know.
There might be something on the HTK mailing list archives - see these links (you need to be logged in to HTK to download these ...):
htk-users.mbox 09-Feb-2007 14:08 20Mhtk-developers.mbox 09-Feb-2007 10:36 2.9M
thanks,
Ken
--- (Edited on 8/23/2008 10:04 pm [GMT-0400] by kmaclean) ---