- Category
- Video & audio
- Status
- In development
- Platforms
- iOS, Android
- Stack
- React Native, Expo, TypeScript, NativeWind, biquad filter chain, LUFS loudness
VoiceCrisp runs a recording through a filter chain — high-pass for rumble, cuts through the muddy range, a presence lift — and then normalises it to a broadcast loudness target. It reports the gain it applied rather than asking you to trust that something happened.
A phone recording of a voice has predictable problems: low-frequency rumble from the desk, a muddy build-up in the low mids, and a level far below what a podcast or a social platform expects. All three are fixable with well-understood filters, and none of them need a server.
The chain
A high-pass removes the rumble, a cut in the low mids clears the mud, and a presence lift brings the consonants forward. Each filter is a biquad with its own tests, so the chain is inspectable rather than a single opaque "enhance" button.
Loudness, not peak
Normalising to peak is the common mistake: it makes the loudest moment hit the ceiling and leaves everything else quiet. VoiceCrisp measures integrated loudness and targets that, because that is the number the platforms use when they turn your audio up or down.
- Tested filter chain
- High-pass, low-mid cut and presence lift, each implemented as a biquad with unit tests.
- Loudness normalisation
- Integrated loudness is measured and matched to a target, rather than normalising to peak.
- Reports what it did
- The measured loudness before and after, and the gain applied, are shown on the result.
- Lifetime unlock
- One purchase covers recordings of any length and removes the ads.
Captures of VoiceCrisp are pending. We would rather show you nothing than a drawing of a screen that was never photographed.