Voiceover and dubbing

flowchart LR
  A[Reviewed words] --> B[Speakers and voices]
  B --> C[Generate speech]
  C --> D[Synchronize and mix]
  D --> F[Export video]
  A --> E[Written captions]
  E --> F

Prepare words before voices

A voiceover project starts with transcription or imported subtitles, then correction and optional translation. Review the selected text before generating speech. Use speaker assignments when the source has several speakers.

Synchronization

Speech must fit the clip. Watch for lines that are much longer in the translation, rapid speaker changes and long pauses. Use the synchronization settings and listen in context. Natural phrasing may require editing rather than simply increasing speed.

Audio choices

Mixed output has source and voice gain settings. Ducking lowers the source while narration is active. Check the whole mix, including quiet lines and background music.

Video and captions

Export a video with no captions, selectable captions or burned captions. An export can reuse finished audio without generating it again. In published 0.12.0, captions follow the text saved with the selected generation run. The development build shown in the screenshots adds a separate choice of current or run-saved text.

The Odra example demonstrates Polish source material with an English Breeze voiceover. See the voiceover tutorial.

Test the difficult passage first

Choose a passage with the longest translated line, a speaker change or an important name. Check the voice and pronunciation, then listen against the original clip. A voice that sounds suitable in isolation may need a different pace or phrasing in the scene.

For several speakers, check the cast assignments before generating the whole plan. Speaker detection and output-voice selection are separate choices. For a cloned voice, use a clean reference in the language you intend to generate and a matching reference transcript.

Review the assembled track

Listen at joins between passages, during pauses and at the end of the clip. Check that synchronization has not made a line unnaturally fast or cut off its final word. Review important lines with the source mix enabled as well as in the generated track alone.

Use an insertion or a music bed only after the spoken timing is working. A sound inserted between lines adds time; an overlay sits on the existing timeline. Source-track ducking and individual sound-cue gain are different settings.

Export checks

• Confirm the intended generation run and written subtitle version.
• Check the source-track and voice gains in the final file.
• Enable selectable captions in a player; burned captions are part of the picture.
• Play the last line and check the video ending. Re-export after changing the audio mix or caption presentation, using the finished audio where appropriate.