Research on source separation and sound localization in real-world environments using informed models and a multi-microphone capture pipeline.
Aude develops an end-to-end audio research workflow: collect synchronized microphone-array data, benchmark state-of-the-art baselines, and train models for robust source separation and localization.
Aude is focused on machine listening in realistic environments where multiple sound sources overlap. The project combines hardware-informed data capture and model design to improve both source separation quality and spatial localization accuracy.
Aude is early-stage and honest about it: a charter, a roadmap, and a hardware plan for machine listening in messy real-world rooms.
Aude is aimed at biologically plausible audio scene analysis: separating overlapping sound sources and locating them spatially in real, reverberant environments rather than clean benchmark recordings. The plan pairs hardware-informed data capture — a small multi-microphone array — with models for source separation and localization, benchmarked against strong baselines.
Aude is run as a disciplined team project, which for research code matters as much as the models:
This is the planning and data-capture stage: the charter, roadmap, and hardware plan exist, and the first runnable baselines are being stood up. I am keeping this page honest — there is no finished architecture to show yet, and I would rather say so than draw one that does not exist.