This is me navigating the Clarity Challenge repository (for the original repo, see https://github.com/claritychallenge) to learn how the tools work, using macOS and VSCode.
The CEC1 challenge is about building hearing aid algorithms to improve speech in noise for hearing aid users. For a brief description of the scenario and the challenge, see this site.
The baseline system includes sample data (rooms, scenes, listeners, and listeners assigned to specific scenes) and walks you through accessing and utilizing the data. The environment is one with stationary noise and simple indoor scenes.
The CEC2 challenge is about improve speech understanding in noise when there is a single target speaker in background noise (multiple sources) in a simulated living room scenario using a range of signal-to-noise ratios and with head movement.
The baseline system includes NAL-R amplification and a compressor, both of which are set to some basic parameters in the config.yaml file. There is no front-end noise cancellation. Processed signals are evaluated with the HASPI metric. The challenge also used a panel of hearing impaired users to assess speech intelligibility.
The CEC3 Challenge builds on the first two challenges by adding real ambisonic room impulse responses, hearing aid recordings, and real dynamic backgrounds. Again, the challenge is to improve speech in noise performance for hearing aids.
The baseline systems are similar to CEC2 but specific for each task. The data is available on MyAirBridge.