Clarity logo
The 3rd Clarity Workshop on Machine Learning Challenges for Hearing Aids (Clarity-CEC2-2022)
EPSRC logo

Results

Results below are for the objective HASPI scores and for provisional subjective listening test scores measured on the evaluation test set.

Note, the subjective listening test scores are provisional. They are based on ASR transcriptions of the listening test responses and are likely to change by a few percent when the tests are rescored by humans. Please make this clear if quoting them. The finalised listening test scores will be released on 1st February.

The table indicates whether systems made explicit use of the head rotation information (Head Rot.) and/or the speaker identity utterances (Spkr. ID.). All systems have been trained only using the official training data.

Rank Team System Spkr. ID. Head Rot. Paper HASPI Score Listener
(Provisional)
T01 E009 Yes [paper] 0.966 93.2
T02 E031 [paper] 0.801 76.5
T03 E008_hr Yes [paper] 0.784 52.6
T03 E008 [paper] 0.777
T04 E037 [paper] 0.775 68.4
T04 E022 [paper] 0.721 65.5
T05 E024_hr Yes [paper] 0.630 44.8
T05 E024 [paper] 0.617
T06 E036 Yes [paper] 0.599 45.6
T06 E038 Yes [paper] 0.554 34.1
T07 E032 Yes [paper] 0.549 35.3
Baseline 0.258 27.0
None 0.172

Prizes


The Hearing Industry Research Consortium prizes for best HASPI score were awarded to

  • 1st place: System E009, Cornell et al, Multi-channel Target Speaker Extraction with Refinement: The WAVLAB Submission to the Second Clarity Enhancement Challenge
  • 2nd place: System E031, Liu and Zhang et al, DRC-NET for The 2nd Clarity Enhancement Challenge
  • 3rd place: System E008_hr, Ouyang et al., The Orka Inc Entry to the 2nd Clarity Enhancement Challenge

Congratulations to the winners!

The Hearing Industry Research Consortium prizes for best subjective listening test performance will be awarded on 1st February when the finalised subjective scores are announced.


[TOP]