$ press play — trust your ears

Hear what CleanTake does

One real voice recording with real room noise. Switch between before and after — same take, processed locally on a regular machine.

Noise type: Steady environmental room noise (fan / AC-type hum) — the exact case CleanTake is built for.
Tested: 2026-09-28 · pipeline: ffmpeg → DeepFilterNet → loudness normalization
Result: noise floor −33dB → −59dB on environmental noise, voice preserved (measured on this sample; your mileage varies with the room).
Formats: MP3 samples; engine tested on WAV/MP3 inputs.

What this demo does NOT prove

Not a café test — overlapping voices are a different problem and CleanTake doesn't solve it (see limits).
Not real-time — this is file processing, not live call cleanup.

How it works under the hood

input.wav ──▶ ffmpeg ──▶ DeepFilterNet ──▶ loudnorm ──▶ output.wav
              decode      denoise (local)      normalize

DeepFilterNet is open source — we credit the original project and don't claim the core tech as ours. CleanTake's value is the packaging: a simple local workflow around a proven engine, free forever. Details in how to use.

Want it on your machine?

CleanTake v2.0.0 is out — Free Forever — every feature free (batch, watch-folder, silence trimming, loudness presets), no license key, no paywall. Download the Windows installer or the portable ZIP. Free, no account, no sign-in.

Download v2.0.0 — free, all features Read the limits first