M-AILABS Speaker Recognition

This is a tool to separate speakers in a list of audio files.

It is quite easy to use. You have a two-step-approach:

Enroll Speakers
Recognize Speakers

Usage

You have two usage models as well: enroll and predict

Enroll

python msr.py -m mymodel.model -c enroll -e data

Predict

python msr.py -m mymodel.model -c predict -p test -l mylogfile.csv

Enrolling Speakers

If you want to enroll speakers, you need to create a directory containing the training data (wav-files). If we call the directory data and we want to identify two different speakers, the structure is like this:

 data
    +--> speaker_1
    |       +---> file1.wav
    |       +---> file2.wav
    |       ...
    +--> speaker_2
    |       +---> file1.wav
    |       ...

Then you can start the tool to enroll them using:

python msr.py -m <model_name> -c enroll -e data

That's all

Predicting

For prediction, equally, you should save all your wav-files you want to analyze into a directory (e.g. 'test') and then you can start prediction.

python msr.py -m <model_name> -c predict -p test [-l mylogfile.csv]

Requirements

You must have python 2.7 and various libraries installed (numpy, scipy). Additionally, you need to install pyssp.

Lastly, please install scikits.talkbox

Optional: fast-gmm and bob (see below)

Installation scikits.talkbox

Install the scikits.talkbox from the contrib-directory using the installation instructions there...

Speed-up

--WORK IN PROGRESS--

You can speed things a little if you compile the library gmm in msrmodel/gmm. Normally, the only thing you have to do is go in there and type:

make

Everything else is already setup. But it may be buggy :-)

--END WORK IN PROGRESS--

You can speed-up things even more if you install bob-packages:

pip install bob.extension bob.blitz bob.core bob.sp bob.ap

That's all folks, hope you have fun...

imdatceleste/speaker-recognition