Spaces:

competitions

/

CryCeleb2023

like 34

Paused

First of all, I want to thank @ubenwa for organizing this fascinating competition.

Since it's still relatively quiet here, I'd like to share my first submission that builds on the baseline using a straightforward Test Time Augmentation (TTA):

Create multiple encoding values by shuffling the order of the audio
Average scores from all possible pairs.
This helped me improve the score from the baseline of 0.3075 to 0.2835.

I hope this helps and have fun!

https://gist.github.com/dienhoa/c391e69128ee5d5b35cc419092167741

[Notes: your score can different than mine because of the random shuffling and I don't choose the seed]