This repository aims to centralize scripts that prepare datasets to be used with pyannote-audio (more precisely, with its pyannote-database dependency).
Currently available :
Each dataset comes with its predefined database.yml, containing pyannote-database protocol(s) with already defined train+dev+test sets for out-of-the-box speaker diarization usage.
How these subsets are defined is entirely configurable.
The python package also contains utilitary cli scripts that can be used to setup the datasets.
- Clone this repository
- Install the python package with
pip install -e . - To setup each dataset, refer to the
README.mdcontained in their respective folder.
Head to the generate_uris.py of the desired dataset, and edit your_subset_creation_logic(), or find the uri-subsets command in the setup.sh script and edit it the same way (some standardization is still needed sorry).
This split can be absolute (= I want X files in subset1 / I want X hours in subset1) or relative (I want X% of the files in subset1 / I want X% of the hours in subset1).
Don't forget to update the database.yml file accordingly.