Voice Editor: review, trim and regenerate a generated narration

Voice Editor is an interface for reviewing and repairing a generated voice-over before it goes into an explainer video. A waveform with a playhead is marked up by act, scene and section; each section carries a WPS score (words per second, for pacing), a VEI expressiveness score graded A to D, and a tone tag such as calm or dramatic. A matrix shows every section at once; selecting one opens its tagged script. From there the reviewer can trim a range by timestamp, regenerate one section or scene through a dialog with a rerollable seed, and A/B-compare old and new takes before keeping or reverting.

The problem and who it is for

The problem it addresses: fixing one flat sentence in a ten-minute narration should not mean regenerating the whole file or opening an audio workstation. It is for the writer or editor finishing a narration.

Stack and status

Stack observed in the source: React 18 with in-browser JSX and a hand-written stylesheet with a mobile layout. Status: an interaction prototype; the text-to-speech vendor is named in the UI, but the regenerate call is simulated and the waveform, spectrogram and every score are generated in the page. It is one of the explainer-video tools listed on the homepage.

Live demo

Prototype status: interaction demo · the sample script, waveform, spectrogram, WPS and VEI scores and every regenerated take are mock data built in the page

Click a section on the waveform or in the matrix to open its detail, then try Trim, Regenerate or A/B Compare. The demo needs JavaScript; it also has a phone layout.