VocalBlade

The problem

Cutting talking-head video by hand is slow and the cuts sound wrong. The goal was editing at the speed of reading, with cuts that are invisible to the ear and the eye.

What I built

• WhisperX transcription with word-level timestamps and speaker diarization (98 languages).
• Audio continuity engine: crossfades, room tone, spectral blending and voice cloning; RIFE/FOMM frame interpolation on the video side.
• Qwen 2.5 7B running locally turns natural-language requests into validated edit operations. Django + GraphQL backend, React + TypeScript frontend.