An implementation of an open-set speaker recognition system for the 1st Multi-target speaker detection and identification Challenge Evaluation (MCE 2018, http://mce2018.org/).
The baseline system of the challenge can be found here: https://github.com/swshon/multi-speakerID. The helper functions used in the implementation of our system are taken from there.
After downloading the data (https://www.kaggle.com/kagglesre/blacklist-speakers-dataset) and installing all required dependencies, you can simply run the script evaluate_system.py which will do everything automatically.
The PLDA script that is used is an adaptation of the MATLAB files of the fastPLDA toolkit (see https://sites.google.com/site/fastplda/) to Python.
When finding this code helpful, or reusing parts of it, a citation is appreciated:
@inproceedings{wilkinghoff2019openspeaker, title={On Open-Set Speaker Identification with I-Vectors}, author={Wilkinghoff, Kevin}, booktitle={The Speaker and Language Recognition Workshop (Odyssey)}, publisher={ISCA}, year={2020} }