TWOEARS / blackboard-system

Two!Ears Auditory Model - Blackboard system module
http://docs.twoears.eu/en/latest/blackboard/
GNU General Public License v2.0
3 stars 2 forks source link

Remove dependency on external GMTK tool #8

Closed hagenw closed 8 years ago

hagenw commented 8 years ago

We should implement GmtkLocationKS that it is purely based on Matlab internal functions and we can get rid of the GMTK dependency.

It might also be a solution to remove GmtkLocationKS if the other localisation knowledge sources are good enough.

/cc: @ningma97 @Hardcorehobel

hagenw commented 8 years ago

For problems with the current GMTK implementation see also https://dev.qu.tu-berlin.de/issues/1900

ningma97 commented 8 years ago

I've added a GmmLocationKS that doesn't require GMTK in the blackboard and an example for showing its performance. Where should I put the trained models though?

hagenw commented 8 years ago

Cool, that was very fast :)

The learned models should be put in the internal database (I'm afraid it would still be a good idea if we could figure out how to make this work on your Mac). At the moment they are stored in this folder: https://dev.qu.tu-berlin.de/projects/twoears-database-internal/repository/revisions/master/show/learned_models/GmtkLocationKS

How large is the learned data? Maybe you can make it available to me and I will put it into the database.

ningma97 commented 8 years ago

OK I've sent it to hagen.wierstorf@tu-berlin.de

Will have a better look at the git-media thing

hagenw commented 8 years ago

I have added the data and the new GmmLocationKS is integrated and working now. I have started a pull request where I will work on the complete removal of the old GmtkLocationKS which will be replaced by GmmLocationKS, see https://github.com/TWOEARS/blackboard-system/pull/11

hagenw commented 8 years ago

The new GmmLocationKS works very nice. I was also able to remove our dependency on GMTK completely. Only the ability to retrain the model is not yet integrated, but this will be done soon and is independent of the task described here, so I will close this issue.