Research
Hearing one sound clearly when many overlap.
My research is on source separation: splitting a mixed recording into the sounds that make it up. It started with speech and is now the core of the project’s acoustic devices.
SPMamba — 2-source to 3-source separation
Presented at UEC ASEAN Seminar and Workshop 2026
Separating several overlapping sources is the same problem the stethoscope and the cough listener face: heart and lung, cough and speech. This study extended the SPMamba model from two to three simultaneous speakers and became the technical base for the project’s acoustic devices.
Research summary. Paper and code are not yet public.
MossFormer 2 — 3-source separation
A second separation architecture trained on the same three-source task, to understand which design is more practical to shrink for small devices.
Research summary. No paper has been published for this work.
From research to devices
- SPMamba / MossFormer 2Edge AI Stethoscope
Separate heart sounds from lung sounds in one recording.
- SPMamba / MossFormer 2Cough Listener
Pull a cough out of speech and room noise.