Own database

59 views
Skip to first unread message

Nathália Batista

unread,
Jan 31, 2018, 1:39:47 PM1/31/18
to bob-devel
Hi all,

I have 960 .wav files to process them in GMM but I'm not understanding how to configure my own database to use in verify.py.

I installed bob using miniconda in my Linux machine.

Is there any file to be generated by bob so I can change it and setup my files on it?

Thank you

Manuel Günther

unread,
Jan 31, 2018, 1:56:45 PM1/31/18
to bob-devel
Dear Nathalia,


Let us know if this is insufficient.

Manuel

Nathália Batista

unread,
Jan 31, 2018, 4:41:41 PM1/31/18
to bob-devel
Hi,

Yes, I read all documentation. The problem is to use bob itself when I do not have the source code to change the configuration. I installed it by conda and created the environment with gmm, spear. Also, I generated the verify.py and I see that is necessary to set each parameter and run it.
But, how or where do I have to change the parameters to setup the new database, preprocessors, extractor, algorithm? When Bob is installed by conda I do not see any folder structure or files that I can change the parameters values and run with verify.py

For example, for the database, I have to create different files following the structure you guys explained in the documentation, but how can I call this by command line once it created?

Sorry, this is my first time using Bob library and i'm quite confused.

Thank you

Nathália Batista

unread,
Jan 31, 2018, 5:48:08 PM1/31/18
to bob-devel
Hi,

I found the bob files in: /miniconda2/envs/bob_py3/lib/python3.6/site-packages/bob/bio

So, now, for me to create a new database configuration I have to create it here: miniconda2/envs/bob_py3/lib/python3.6/site-packages/bob/bio/spear/config/database

Is that right?

Thanks

Manuel Günther

unread,
Jan 31, 2018, 6:38:10 PM1/31/18
to bob-devel
No. You can basically create configuration files wherever you want, and pass those files on the command line to verify.py. The verify.py script can take configurations in three different ways:

1. as path to configuration files
2. as registered resources
3. (rarely used) by instantiating classes on the command line


The recommended way is to write a single configuration file that includes the complete configuration of your experiment, including the configuration of your database. This configuration file can be anywhere, but usually people keep the configuration file in the current directory -- do not write anything into miniconda2/envs/bob_py3/lib/python3.6/site-packages/bob.

I hope this helps.

Manuel

Amir Mohammadi

unread,
Feb 1, 2018, 5:00:25 AM2/1/18
to bob-...@googlegroups.com
I think the software for our ICML paper:
https://gitlab.idiap.ch/bob/bob.paper.icml2017
is a great example on how to extend bob.bio.base
through configuration files.

Amir

--
-- You received this message because you are subscribed to the Google Groups bob-devel group. To post to this group, send email to bob-...@googlegroups.com. To unsubscribe from this group, send email to bob-devel+...@googlegroups.com. For more options, visit this group at https://groups.google.com/d/forum/bob-devel or directly the project website at http://idiap.github.com/bob/
---
You received this message because you are subscribed to the Google Groups "bob-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bob-devel+...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Nathália Batista

unread,
Feb 1, 2018, 5:35:01 PM2/1/18
to bob-devel
Thank you all for clarifying that.

Nathália Batista

unread,
Feb 2, 2018, 5:32:34 PM2/2/18
to bob-devel
Hi guys,

I do have another question. How do we determine model_id and client_id for dev (eval) ? Do they depend on the number of files that I have in the database?
I'm trying to understand this based on Timit example in the bob source.

Thank you

Tiago Freitas Pereira

unread,
Feb 3, 2018, 1:54:53 AM2/3/18
to bob-...@googlegroups.com
Hi Nathália,

Just clarifying, model_id is a field that groups all the files used to create a particular model (speaker model in you case).
One model can be composed by several audio files.
The client_id is the id of a particular subject.

Although this is not common, one client can have more than one model.
Most our databases model_id = client_id, meaning that we have only one model per client.
If you database fits this case, you can safely do client_id = model_id

I hope it helps

Cheers


--
-- You received this message because you are subscribed to the Google Groups bob-devel group. To post to this group, send email to bob-...@googlegroups.com. To unsubscribe from this group, send email to bob-devel+unsubscribe@googlegroups.com. For more options, visit this group at https://groups.google.com/d/forum/bob-devel or directly the project website at http://idiap.github.com/bob/

---
You received this message because you are subscribed to the Google Groups "bob-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bob-devel+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.



--
Tiago

Nathália Batista

unread,
Feb 3, 2018, 5:29:01 AM2/3/18
to bob-devel
Hi Tiago,

Yes, It helped a lot.

Thank tou
Tiago
Reply all
Reply to author
Forward
0 new messages