How to use Bob's i-vectors functions in matlab environment?

[Jinfang Wang]

Hi,

I would like to call the function extacting i-vectors implemented by Bob in matlab code environment because there are much work having been done with matlab program. I guess that there might be two ways,

1) In matlab environment, call Python Interpreter and 'import bob.bio.gmm' as well as related python codes. Maybe this can be regarded as 'matlab-python hybrid programming'. However, I amn't sure whether this is feasible.

2) In matlab environment, imbed 'verify.py ...' related command lines in the form of linux commands in matlab codes directly. I also can't be sure its availability.

My questions are,

1> Could you tell me whether two ways are available?

2> How to implement them with the concrete procedures?

3> If both of them are practicable, which is more efficient? Which one do you suggest me to choose?

Thank you for your warmhearted help.

Jinfang

[Manuel Günther]

Dear Jinfang,

I don't think that it is possible to call Python from within Matlab, but I am not a Matlab programmer. 

On the other hand, I know that there are some wrappers to call Matlab code from within Python, such as pymatlab. I have used this wrapper once to use Matlab code (that I got from someone else) inside my Python environment. Maybe you want to explore this option.

Anyways, even if there is a possibility to use Python from within Matlab, bob.bio.gmm might not be the best starting point. While bob.bio.gmm provides code to extract I-Vectors, it does so within the Biometrics framework. Maybe, it would be easier to interface with the toolbox, which implements I-Vectors directly, which would be bob.learn.em (AFAIK). You might want to have a look into the implementation inside bob.bio.gmm to see, which functions are called for I-Vector extraction.

If there is no direct interface between Matlab and Python, you are still not lost. You might perform the I-Vector extraction with Python, write the I-Vectors to an HDF5 file, and read HDF5 with Matlab.

Finally, the better option would be to drop Matlab and completely switch to Python. I am not sure if this is feasible, though.

Best regards

Manuel