Adjusting silence models

[aolney]

Thanks for a great github project. I'm trying to adjust the silence models and haven't been able to come up with much except this:

pmsseg=./$datadir/$show.pms.seg
java -Xmx$mem -classpath "$LOCALCLASSPATH" fr.lium.spkDiarization.programs.MDecode  --help  --fInputDesc=$fDescD --fInputMask=$features --sInputMask=./$datadir/%s.i.seg --sOutputMask=$pmsseg --dPenality=10,10,50 --tInputMask=$pmsgmm $show

where

Penality set penalties for the i to j states 10 for the first and second model corresponding to boad/narrowband silence 50 for the other jingle speech (f0 f2 f3 fx), jingle and music

So I have 2 questions:

1. How to determine these parameters
2. Are there any other options for adjusting the silence/non speech models

[StevenLOL]

Hi, I am very sorry, I am not clear about how to determine those parameters.

Maybe you can contact the LIUM team

Do let me know if you get the answers.

Thanks.

[aolney]

Will do thanks

[libeppi]

i wasn't able to find any answer even grubbing among source in the .jar however i have another (quite related) question: let's say that i record 10 seconds, 3 of speech and the rest of light noisy background. pms analysis is telling me that silence occoured in a certain range with label ''iS'' How come in the next step (see below) silence gets clustered as a new and unique speaker? Any tips?

echo filter spk segmentation according pms segmentation
fltseg=$datadir/$show.flt.$h.seg
java -Xmx$mem -classpath "$LOCALCLASSPATH" fr.lium.spkDiarization.tools.SFilter  --fInputDesc=$fDescD --fInputMask=$features --fltSegMinLenSpeech=150 --fltSegMinLenSil=25 --sFilterClusterName=j --fltSegPadding=25 --sFilterMask=$pmsseg --sInputMask=$datadir/%s.d.$h.seg --sOutputMask=$fltseg $show

[aolney]

That's not what happens to me, but thanks for pointing out the iS code -- I hadn't noticed that.

For my data, using this github repo, the iS disappears at the flt stage.