VoiceCut

Open-source audio and video editing

VoiceCut — automatic narration editing for content creators.

Remove false starts, retakes, corrections, and recording directions from audio or video while preserving the speaker's real voice.

Explore examples View on GitHub

Real examples

Compare the source and final edit

These files were produced by the normal VoiceCut pipeline without manual repair. Starting one player automatically pauses the others.

English narration

False starts become one clean take

Original

38.00 s

VoiceCut

27.34 s

The abandoned opening and repeated phrase are removed; the complete intended take remains.

Русская речь

Повторы превращаются в связное объяснение

Оригинал

35,07 с

VoiceCut

21,68 с

Неудачные попытки и повторённые фразы удалены, а финальное объяснение сохранено.

Video editing

The picture follows the selected speech

Original

30.25 s

VoiceCut

23.78 s

VoiceCut applies direct visual cuts at the selected speech ranges—no frozen frames and no artificial video pauses.

How it works

One semantic plan. One final render.

VoiceCut first determines what the speaker intended to keep, then resolves safe source boundaries before rendering once from the original recording.

VoiceCut pipeline from canonical audio or video through Whisper transcription, semantic planning, source grounding, completeness validation, MFA phone alignment, pause and ambience planning, an immutable boundary plan, and one final render
The planner chooses source occurrences; MFA supplies phone-safe coordinates; the immutable plan is rendered once from the canonical source. Open the diagram full size. Read the technical overview.

Quick start

Run VoiceCut with one command.

voicecut recording.wav
Install from GitHub