Competitive baselines and leaderboard availability


Firstly thank you for organising the competition, anything looking to progress the state of audio ML is fantastic.

Firstly, I was wondering if there were any good resources to get started in the field of generic audio embeddings. I’ve come across the OpenL3 and PANNs repos, but haven’t found a good paper comparing SOTA embeddings for many tasks.

Secondly, I believe the first round had already completed (and I’ll miss the August round too). Is it possible to make the leaderboard public? Given the limited opportunity to test performance (compared to Kaggle / competition platforms) it would be good to know I’m competitive on the open tasks before submitting.

Thanks in advance!

1 Like


Actually, we are aware of no broad-scale audio embedding benchmarks, hence this competition. But here are some papers besides OpenL3 and PANNs to get started in generic audio embeddings:

  • COLA audio embeddings
  • YAMNet
  • VGGish
  • SoundNet
  • NSynth Wavenet and Wavenet like models
  • Google LEAF audio frontend
  • Wav2Vec family
  • CPC audio
  • PASE audio
  • MockingJay
  • Audio ALBERT
  • NPC audio
  • APC audio
  • BYOL-A

No leaderboard has been posted yet, but the first leaderboard will be up in two weeks. Please see recent updates about rolling leaderboard submissions