Remove silences from a recording automatically

Automatic silence cutting is the highest-leverage edit in spoken video, and the one most likely to overreach. The beat before a punchline, the pause where someone is thinking, the breath that makes a sentence land — cut all of those and the result is tighter and worse.

So the automatic pass is a starting point, not a verdict. Every cut is a region you can see, drag, or delete, reinstating a pause takes a click, and the preview plays the edit as it stands.

Removing dead air is rarely the only thing a recording needs. In the same session Rescript will strip every filler word, let you delete whole sentences by selecting them in the transcript, label who's speaking, and respeak a line you fumbled in that speaker's own voice.

Because all of it happens against one transcript with word-level timings, the cuts compose cleanly — remove silences, remove fillers, then delete the tangent in the middle, and the export is a single word-accurate render of what's left.

Transcription runs through WebGPU on your own GPU, with a WASM fallback. Diarization runs locally. The export runs through ffmpeg compiled to WebAssembly on your CPU. Once the model files have downloaded once, the network is optional.

So there's no upload wait before you can start, no queue, no per-hour billing, and no limit on how long the recording can be beyond what your machine can hold.