Understanding speech in noisy, everyday environments remains a major challenge for hearing-aid users. The Clarity Project, a six-year UK research programme, addresses this by running a series of international machine learning challenges on speech intelligibility enhancement and prediction. The 3rd Clarity Prediction Challenge (CPC3) built on CPC1 (simple, stationary scenes) and CPC2 (multiple interferers with head movements) by introducing fully dynamic listening environments with real backgrounds and measured hearing-aid signals. Participants were asked to predict, from hearing-aid outputs and listener audiograms, the percentage of words a hearing-impaired listener would correctly recognize. Systems were evaluated on root mean squared error (RMSE) and correlation across thousands of listener–signal pairs. CPC3 attracted 21 submissions from 14 teams worldwide, spanning intrusive and non-intrusive approaches based on speech foundation models. The winning system (E025) achieved an evaluation RMSE of 24.98 and correlation of 0.80, significantly outperforming strong baselines and marking clear progress beyond CPC1 and CPC2.

@inproceedings{barker25_clarity,
  title     = {{The 3rd Clarity Prediction Challenge: A machine learning challenge for hearing aid intelligibility prediction}},
  author    = {Jon Barker and Michael A. Akeroyd and Trevor J. Cox and John F. Culling and Jennifer Firth and Simone Graetzer and Graham Naylor},
  year      = {2025},
  booktitle = {{The 6th Clarity Workshop on Improving Speech-in-Noise for Hearing Devices (Clarity-2025)}},
}