stevenlol / lium

Scripts for LIUM SpkDiarization tools

LIUM

A demo on LIUM ILP clustering

LIUM_SpkDiarization is a software dedicated to speaker diarization (ie speaker segmentation and clustering), developed by Laboratoire d'Informatique de l'Université du Maine (LIUM - EA 4023).

---

Homepage: http://www-lium.univ-lemans.fr/diarization/doku.php/welcome

Download http://www-lium.univ-lemans.fr/diarization/doku.php/download

Related project: https://github.com/bbcrd/diarize-jruby

Prerequisites:

1)glpk for ILP clustering:

sudo apt-get install glpk

or

sudo apt-get install glpk-utils

http://www.gnu.org/software/glpk/

2)UBM and models http://www-lium.univ-lemans.fr/diarization/lib/exe/fetch.php/data_ilp.tgz

After download LIUM, UBM and other models, put them in this folder, you should have following file structure, then just $./go.sh, output segmetns will be in test_out folder

./
|-- LIUM_SpkDiarization-8.4.1.jar
|-- go.sh
|-- ilp_diarization2.sh
|-- LICENSE
|-- README.md
|-- mat
|   |-- wld.efn.xml
|   |-- wld.mahanalobis.mat
|   `-- wld.tv.mat
|-- models
|   |-- gender.gmms
|   |-- s.gmms
|   |-- sms.gmms
|   `-- ubm.gmm
|-- test_out
|-- test_wav
|   `-- t001.wav
`-- ubm
    `-- wld.gmm

Details are in ilp_diarization2.sh

Known issues:

1. #1 LIUM works fine with JDK 1.7 but has some problem with JDK 1.8

Please check out this project ,it works with JDK1.8 .

Update:

LIUM=Laboratoire d'Informatique de l'Université du MaineLaboratoire d'Informatique de l'Université du Maine

http://www-lium.univ-lemans.fr/

Speaker diarization

Speaker diarization is one research topic in audio/speech processing. It is refer to unsupervised cluster audio sequences into speaker groups.

lium's Issues

Limiting the number of speakers

Hey
Is there any way of limiting the number of final clusters detected to 2? Since I am using LIUM_diarization on a one on one conversation diarization.

Adjusting silence models

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

ILP clustering - JDK 1.8

Hello, student here, your script is great! It really helped me better understand speech segmentation.
I'm able to run it till ILP-clustering (last step), when i try to go ahead this is what i get:

11:42.567 TrainIVectorOrTINFO | Compute statistics {make() / 1}
11:42.731 TrainIVectorOrTINFO | Train i-vector {make() / 1}
11:42.732 TotalVariabilitINFO | sent #: 0 {estimateL() / 1}
Exception in thread "main" com.thoughtworks.xstream.converters.ConversionException: -1 : -1
---- Debugging information ----
message : -1
cause-exception : java.lang.ArrayIndexOutOfBoundsException
cause-message : -1
class : fr.lium.spkDiarization.libModel.gaussian.FullGaussian
required-type : fr.lium.spkDiarization.libModel.gaussian.FullGaussian
converter-type : com.thoughtworks.xstream.converters.reflection.ReflectionConverter
path : /fr.lium.spkDiarization.libModel.ivector.EigenFactorRadialList/globalMeanCovariance/gender
class[1] : fr.lium.spkDiarization.libModel.ivector.EigenFactorRadialList
version : 8.4.1 09/12/2013 10:48 AM

i didn't install glpk via apt-get because as of today (UBUNTU 14.04) it is not working. I tried manually with the old fashion way but i don't know if everything went ok.

Any tips?
Thank you
Giuseppe

train ivector

Recently, I am doing speaker diarization with LIUM, After I trained Total variability matrix and i-vector, when train EFR normalization, I met some problem about Eigenvalue Decomposition failed, because org.ejml.simple.simpleEVD.(Unknown Source).Do you know what happened about it.

Where can I download the data_ilp.tgz file now

Hi
This link(http://www-lium.univ-lemans.fr/diarization/lib/exe/fetch.php/data_ilp.tgz) is now invalid? and UBM and models How was them trained?
thanks！