Audio Noise Remover
Remove background noise from audio files locally with the GTCRN model.
How it works
- 1Select an audio file — it is decoded inside your browser, never uploaded.
- 2Audio is resampled to 16 kHz mono and split into 32 ms overlapping frames.
- 3The GTCRN model (~535 KB, MIT licensed) enhances each frame with recurrent state, running on ONNX Runtime WASM.
- 4The enhanced frames are reassembled into a WAV you can play and download.
FAQ
Is my audio uploaded anywhere?
No. Decoding and enhancement run entirely inside your browser. Your audio never leaves your device.
Which kinds of noise does it handle?
GTCRN is trained on the DNS challenge dataset — steady broadband hiss, fan noise, hum and general background babble. Sudden impulsive noises (claps, clicks) are less affected.
What audio length is supported?
Up to 10 minutes per file in the browser. The model processes the clip frame by frame, so longer files just take proportionally longer.
Why does the output sound slightly different from the original?
The model re-synthesises the spectrum, which can subtly change timbre while removing noise. Speech usually comes out cleaner; heavily corrupted sections may sound muffled.