SafeANC
Noise cancellation that gets out of the way when an ambulance is coming.
- Timeline
- Jan 2026 – present · Building
- Role
- Solo
- Modality
- Audio
- Evaluated on
- <50 ms latency budget · adversarial set
- Hard latency budget
- <50 ms
- Deployment modes
- 2
- Evaluation set
- Adversarial
- Status
- Building
The problem
Active noise cancellation is very good at removing exactly the sounds you sometimes need to hear. Sirens are the clearest case: a system doing its job perfectly can suppress the one sound that matters for your safety.
What I'm building
A real-time audio ML system that detects emergency sirens (ambulance, police, fire) and selectively bypasses noise cancellation for them while leaving everyday noise suppression intact. The architecture is modular and deployable two ways: as a drop-in safety layer over an existing ANC system via audio passthrough, or as a full end-to-end ANC pipeline with detection built in. I am crowdsourcing acoustic clips from vehicle interiors, sidewalks, and traffic to cover the hard edge cases, with a data pipeline that standardizes variable-length, variable-quality recordings into a consistent evaluation set.
How I know it's working
The whole design is bounded by a <50 ms latency budget for real-time inference on embedded hardware, so accuracy that arrives late is not accuracy. The evaluation set is built deliberately adversarial: sirens at distance, sirens under traffic noise, and sounds that resemble sirens but are not. A safety system's only interesting failure mode is the miss, so that is what the edge-case collection targets.