multi-speakerID baseline

This is python implementation of multi-target speaker recognition based on i-vector feature. This is also baseline system of the first Multi-target speaker detection and identification Challenge Evaluation (MCE 2018, http://www.mce2018.org )

Dataset

You can download i-vector dataset here

https://www.kaggle.com/kagglesre/blacklist-speakers-dataset

After download, extract into data folder

System flow

Performance

If you run the code like

python mce2018_baseline_test.py

you will see the performance on top-S and top-1 detector as below :

Dev set score using train set :
Top S detector EER is 2.01%
Top 1 detector EER is 12.26% (Total confusion error is 444)

Test set score using train set:
Top S detector EER is 7.52%
Top 1 detector EER is 16.61% (Total confusion error is 579)

Test set score using train + dev set:
Top S detector EER is 6.24%
Top 1 detector EER is 11.24% (Total confusion error is 369)

And the code also generate example submission file with name "teamname_fixed_primary.csv" and the format are [test utterance ID],[score],[Closest blacklist speaker ID] per each files. For example

pnah_431154,0.36613864,33762391
qtxw_470243,0.39585015,60587769
....

Reference

If you want to check detailed information, please read below

Suwon Shon, Najim Dehak, Douglas Reynolds, James Glass,
"MCE 2018: The 1st Multi-target speaker detection and identification Challenge Evaluation (MCE) Plan, Dataset and Baseline System”
https://arxiv.org/abs/1807.06663

Question

Please email to mce organizer if you have question. mce@lists.csail.mit.edu or swshon@mit.edu