How to start?

[MSAlghamdi]

I compiled Alize LIA_RAL successfully --with-spro[=path] and, before that, compiled SPro successfully --with-sphere[=path] . Now, I'm trying to test it with a dataset. I tried to find a guideline or a set of instructions that explain how to run it but I couldn’t find any source online. On the Alize webpage It clearly says “the documentation for the library APIs is in dire need of an update, and is not therefore available for the current version of ALIZÉ”!

Could you please help me to start using LIA_RAL?

[tevamerlin]

Hi,

Have you tried the tutorials available on the ALIZÉ webpage? Indeed, most uses of LIA_RAL are not through direct calls to the API, but by combining the various binaries to build a speaker recognition system. The tutorials do just that, using shell scripts. The scripts are commented to highlight the various steps.
Just replace the executable binaries in the tutorials with the ones you just compiled, in order to use the latest version of LIA_RAL.

[MSAlghamdi]

^
Thank you very much! That was helpful, but got some issues that I couldn't solve when I ran the 03_RUN_gmm-ubm.sh script. It's my master's thesis and I need to run the kit in few weeks!

The 1st issue is the channelCompensation parameter. I don't know in which config file I should set it. The other one is about FileReader.cpp. There're more errors related with this file.

It shows as following:

# sh 03_RUN_gmm-ubm.sh
Train Universal Background Model by EM algorithm
		done, see log/TrainWorld.log for details
Train Speaker dependent GMMs
		done, see log/TrainTarget.cfg for details
Compute Likelihood
		done, see log/ComputeTest.cfg
Train Impostor dependent GMMs for T-norm
		done, see log/ComputeTest_tnorm.log
Compute Z-norm scores
		done, see log/ComputeTest_znorm.log
Compute T-norm scores
		done, see log/ComputeTest_tnorm.log
Compute ZT-norm scores
		done, see log/ComputeTest_ztnorm.log
Apply Z-norm
Begin world model training
TrainTarget [ ConfigCheckException 0x2180a70 ]
  message   = "Parameter 'channelCompensation' mandatory in the config"
  source file = ConfigChecker.cpp
  line number = 129
--help           Show this help
--version        Show version information
--normalizeModel if set to true,  normalize the world (at each iteration) <BOOLEAN>

TrainTarget [ ConfigCheckException 0x1bbda60 ]
  message   = "Parameter 'channelCompensation' mandatory in the config"
  source file = ConfigChecker.cpp
  line number = 129
--help           Show this help
--version        Show version information
--normalizeModel if set to true,  normalize the world (at each iteration) <BOOLEAN>

		done, see log/ComputeNorm_znorm.log
Apply T-norm
		done, see log/ComputeNorm_tnorm.log
Apply ZT-norm
		done, see log/ComputeNorm_ztnorm.log
(SegTools) >> Proceeding label file reading for [xaaf] <<
root@88162bef7632:/root/ipython/alize_proj/LIA_RAL# [ ConfigCheckException 0x26a5800 ]
  message   = "Parameter 'channelCompensation' mandatory in the config"
  source file = ConfigChecker.cpp
  line number = 129
--help    Show this help
--version Show version information

(ComputeTest) No Channel Compensation
(ComputeTest) No Channel Compensation
Compute Test, Decime mode[1] File Mode- 1 LLR by file
Compute Test, Decime mode[1] File Mode- 1 LLR by file
[ FileNotFoundException 0x1dec130 ]
  message   = ""
  source file = FileReader.cpp
  line number = 101
  fileName =  ndx/computetest_gmm_target-seg.ndx
[ FileNotFoundException 0x2063130 ]
  message   = ""
  source file = FileReader.cpp
  line number = 101
  fileName =  ndx/computetest_gmm_target-imp.ndx
[ FileNotFoundException 0xcce300 ]
  message   = ""
  source file = FileReader.cpp
  line number = 101
  fileName =  res/target-seg_gmm.res

[utlandr]

I would like to add that I am also having issues with the channelCompensation parameter. It doesn't appear to be a required parameter in the various configs I have seen. @MSAlghamdi , were you able to get around this?

[MSAlghamdi]

If you use LFA, JFA or NAP, you need to assign channelCompensation parameter in the config file. In LFA, JFA and NAP codes there are methods (that I couldn't remember) deal with channel variation. When you run TrainWorld and TrainTarget it's required (by main function) to determine if you are using LFA, JFA or NAP by assigning channelCompensation parameter in the config. file to the one you used. If you're using GMM_UBG with MAP adaptation only, you can set channelCompensation as none.

[hananeam1991]

mv TurnDetection ../../../bin
make[3] : on quitte le répertoire « /home/hanane/Bureau/PACK-ALIZE/LIA_RAL-master/LIA_SpkSeg/TurnDetection/src »
make[3] : on entre dans le répertoire « /home/hanane/Bureau/PACK-ALIZE/LIA_RAL-master/LIA_SpkSeg/TurnDetection »
make[3]: rien à faire pour « all-am ».
make[3] : on quitte le répertoire « /home/hanane/Bureau/PACK-ALIZE/LIA_RAL-master/LIA_SpkSeg/TurnDetection »
make[2] : on quitte le répertoire « /home/hanane/Bureau/PACK-ALIZE/LIA_RAL-master/LIA_SpkSeg/TurnDetection »
make[2] : on entre dans le répertoire « /home/hanane/Bureau/PACK-ALIZE/LIA_RAL-master/LIA_SpkSeg »
make[2]: rien à faire pour « all-am ».
make[2] : on quitte le répertoire « /home/hanane/Bureau/PACK-ALIZE/LIA_RAL-master/LIA_SpkSeg »
make[1] : on quitte le répertoire « /home/hanane/Bureau/PACK-ALIZE/LIA_RAL-master/LIA_SpkSeg »
make[1] : on entre dans le répertoire « /home/hanane/Bureau/PACK-ALIZE/LIA_RAL-master »
make[1]: rien à faire pour « all-am ».
make[1] : on quitte le répertoire « /home/hanane/Bureau/PACK-ALIZE/LIA_RAL-master »
i have this error