The speech diarization project obtains recorded speech with two speakers (each speaking into their own microphone) and returns an output file with the time intervals that they were speaking.

matplotlib
scipy
sounddevice

The training data should be placed into the media folder:
deep-diarization/media/TextGrids/
deep-diarization/media/Sound Files/
etc. These files are listed in the git ignore and should never be committed into the repository because they are huge.

To run the training process, open the main.py file, and adjust the parameters as needed and run the script. Eventually, all parameters will be adjustable from the command line.

To run the diarization method, set the training_flag to False. This will skip the training process and write the textgrid files to disk.

The folder diarization_methods should be the place to add new diarization methods. Each method should be a class that inherits from the DiarizationBaseClass. An example class, that does nothing but print out a statement in each method, is provided as a reference.

After a diarization method has been created, change the diarization_method parameter found in main.py to use your method.

There are also class and method comments inside of DiarizationBaseClass that will be worth reading, as it explains much more about how to use it.

Unit tests are a great way to help debug. Writing a unit test is easy with the nosetests framework.

All you do is:

1. Create a TestSomething.py (must start with 'test') in the tests folder.
2. Add methods into it to test each function of your class or method. Test methods names should start with 'test'.
3. Test methods should also make liberal use of assert statements. These are used to say 'If this statement is false, the test should fail'. eg assert x == 4
4. Then, in the terminal, cd to the tests folder and run nosetests. You might have to install it first, use pip or apt. nosetests with then go through all the tests within the tests folder and show you which ones pass and which ones fail.
5. Eat a taco and enjoy how easy that is.

• Clint Ferrin
• Madi Mickelsen
• Daniel Mortensen
• Josh Vanfleet