CallCleaner
A lightweight bioacoustic denoising model that transfers capability via knowledge distillation from the general-purpose audio separation foundation model SAM-Audio, with approximately 19 million parameters, requiring no detector, and capable of running in real time on a CPU.
Timeline (last 90 days)
CallCleaner is a denoiser with only 19 million parameters, detector-free, and runs at approximately 90 times real time on a laptop CPU.
The researchers used knowledge distillation to transfer the capabilities of SAM-Audio to a small student model, CallCleaner.
CallCleaner leaves 11 dB less noise around calls than Biodenoising, the leading publicly available neural bioacoustic denoiser.
Among fast, CPU-available systems, CallCleaner occupies an advantageous position on the Pareto frontier of noise removal versus call fidelity, and requires no detector.
The approach used in this paper requires only lightweight temporal hints to guide the separation model to focus on time periods when calls occur, without requiring precise frame-by-frame masks, thus reducing dependence on detection accuracy and keeping the system detector-free and lightweight.
CallCleaner is designed for single-vocalizer recordings and supports fast, private local denoising without specialized hardware.
The researchers have open-sourced the code and weights of CallCleaner.
All Facts (7)
CallCleaner is a denoiser with only 19 million parameters, detector-free, and runs at approximately 90 times real time on a laptop CPU.
50%UnverifiedThe researchers used knowledge distillation to transfer the capabilities of SAM-Audio to a small student model, CallCleaner.
50%UnverifiedCallCleaner leaves 11 dB less noise around calls than Biodenoising, the leading publicly available neural bioacoustic denoiser.
50%UnverifiedAmong fast, CPU-available systems, CallCleaner occupies an advantageous position on the Pareto frontier of noise removal versus call fidelity, and requires no detector.
50%UnverifiedThe approach used in this paper requires only lightweight temporal hints to guide the separation model to focus on time periods when calls occur, without requiring precise frame-by-frame masks, thus reducing dependence on detection accuracy and keeping the system detector-free and lightweight.
50%UnverifiedCallCleaner is designed for single-vocalizer recordings and supports fast, private local denoising without specialized hardware.
50%UnverifiedThe researchers have open-sourced the code and weights of CallCleaner.
50%