How the denoiser works
- Model
- DeepFilterNet3, an open-source speech enhancement model (MIT / Apache 2.0). It works on 48 kHz full-band audio in 10 ms frames.
- Where it runs
- In your browser, as WebAssembly. The model downloads once and stays in your browser's cache; your file is never uploaded.
- Formats
- FFmpeg, also running in your browser, converts MP3, WAV, M4A, FLAC, videos and more to 48 kHz before denoising, then encodes the result.
- Long files
- Long recordings are denoised in cross-faded segments, so memory use stays low, even on a phone. On Pro and Lifetime, files over 5 minutes use several CPU cores.
- Works offline?
- No. The page and, the first time, the model need a connection; the denoising itself happens on your device.
Why You Need Audio Denoising
Hiss, hum, and ambient noise degrade audio quality in recordings. Professional denoising tools are expensive and complex. AudioClean brings studio-grade AI denoising to your browser for free.
How AudioClean Solves It
Three simple steps to clean audio.
Step 1
Upload
Drag and drop your file. Supports 25 audio and video formats.
Step 2
AI Cleans
AI identifies and removes noise while preserving voice clarity.
Step 3
Download
Preview with A/B comparison, then export.
Recommended Settings
Level
Medium
Common Use Cases
Removing hiss from microphone recordings
Cleaning hum from electrical interference
Denoising audio tracks for video production
Restoring clarity to old or degraded recordings
Frequently Asked Questions
They're the same thing. Denoising (or noise reduction) uses AI to identify and remove unwanted sound while preserving the desired audio signal like speech or music.
AudioClean uses DeepFilterNet3, a state-of-the-art deep learning model for real-time audio denoising. It processes at 48kHz with 10ms frame latency.
AudioClean is optimized for speech and voice recordings. It may work on music, but results vary. Use the "Light" level to minimize artifacts on music tracks.