Pre-Announcing the 3rd Clarity Speech Enhancement Challenge for Hearing Aid Signal Processing
Trevor Cox
The third Clarity Enhancement Challenge for Hearing Aid Signal Processing (CEC3) is opening soon.
Important Dates
2nd April - Challenge Launch.
25th July - Evaluation data released
2nd Sept - First round submission deadline (objective)
Sept-Nov - Listening test evaluation period.
Dec - Results announced at a Clarity Challenge online workshop and prizes awarded
Background
Clarity is organising a series of machine learning challenges to advance hearing aid speech signal processing.
The 3rd Clarity Speech Enhancement Challenge
Our previous rounds have focused on domestic environments using simulated scenes. This new challenge will have separate tracks to encourage progress in two complementary directions: using signals from real hearing aid recordings and using real dynamic scenes captured using Ambisonic recordings.
As with previous challenges, we will be evaluating the best entries using a panel of hearing-impaired listeners. There will be cash prizes available to the top system(s), courtesy of the Hearing Industry Research Consortium.
Track 1 - Real hearing aid recordings
In the previous challenges, CEC1 and 2, all the data relied on simulations. In CEC3, we will be providing data recorded on hearing-aid shells worn by listeners in rooms. So this track is testing your ability to make a system that can generalise to more realistic real-world scenarios. The scenario is domestic living room listening as in the case of CEC1 and 2. Participants will need to handle the effects of real room acoustics, hearing aid user head motion and hearing aid microphone effects.
Track 2: Real dynamic background noises
In our previous challenges, the interfering signals have been static and carefully controlled. In CEC3 Track 2, we will use naturally occurring, dynamic noise backgrounds. We are collecting a dataset of 64-channel Ambisonic recordings from places that hearing-impaired listeners find challenging. These include train stations, roadsides and large social gatherings (i.e., the 'cocktail party' scenario). Using these recordings and measured impulse responses, we will create a dataset of hearing aid input signals that have target sentences in dynamic background noise.
Evaluation
For both tracks, we will be providing standard training, development and evaluation datasets. The training and development datasets will be released at the start of the challenge. The evaluation dataset will be released shortly before the submission deadline without reference signals. Participants will then be asked to submit their processed signals for remote evaluation. Subsequently, up to twenty of the most promising systems (up to 10 for each track) will be evaluated by a panel of listeners. We will provide a baseline system that provides a standard hearing aid amplification stage, so that teams can choose to focus on individual components or to develop their own complete pipelines.
What will be provided
Evaluation of the best entries by a panel of hearing-impaired listeners.
Speech + interferer scenes for training and evaluation.
Listener characterisations including audiograms and speech-in-noise testing.
Software including tools for generating training data, a baseline hearing aid algorithm, a baseline model of hearing impairment, and a binaural objective intelligibility measure.
Challenge and workshop participants will be invited to contribute to a journal special issue on the topic of Machine Learning for Hearing Aid Processing, which will be announced next year.
For further information
If you are interested in participating and wish to receive further information, please sign up to the Clarity Challenge Google Group at https://groups.google.com/g/clarity-challenge
If you have questions, contact us directly at claritychallengecontact@gmail.com
Organisers (alphabetical)
Michael A. Akeroyd, Hearing Sciences, School of Medicine, University of Nottingham
Jon Barker, Department of Computer Science, University of Sheffield
Trevor J. Cox, Acoustics Research Centre, University of Salford
John F. Culling, School of Psychology, Cardiff University
Jenny Firth, Hearing Sciences, School of Medicine, University of Nottingham
Simone Graetzer, Acoustics Research Centre, University of Salford
Graham Naylor, Hearing Sciences, School of Medicine, University of Nottingham
Jianyuan Sun, Department of Computer Science, University of Sheffield
Funded by the Engineering and Physical Sciences Research Council (EPSRC), UK
Supported by RNID, Hearing Industry Research Consortium, Amazon TTS Research