How it works
Detection from the signal, not the clock.
Autocut decodes your video's audio in the browser, measures loudness in 20ms windows, and separates speech from pauses with controls you can audit.
- 1 · Decode locally
Web Audio decodes the track on-device. A downsampled peak map renders the waveform; nothing is uploaded.
- 2 · Measure loudness
Each window becomes an RMS level in dB. The silence threshold (−50 to −20 dB) splits candidate quiet frames from speech.
- 3 · Confirm with hysteresis
A gap must stay quiet ~120ms to open and stay loud ~80ms to close — so breaths and plosives don't shatter into cuts.
- 4 · Apply gap + padding
Gaps shorter than minimum silence are discarded. Keep before/after padding pulls cut edges back so word onsets survive.
- 5 · You review, then export
Every cut lists timestamps, duration, and confidence. Toggle keep/remove, preview the tightened cut, export keep-ranges via real-time capture.