July 2026
- Clone method is now a preference: choose how a speaker's voice reference is applied instead of taking the built-in default.
- Long videos resume instead of restarting. A retry now continues from the last completed point rather than re-rendering the whole timeline, so a failure near the end no longer costs you the whole job again.
- Retry is available on far more failures. Errors you can actually resolve — including running out of balance mid-job — now unblock the retry button once you have fixed the cause, instead of leaving the task permanently dead.
- Videos that finished with a degraded video track can now have the video leg regenerated on its own, without redoing the audio.
- Translation quality: prompts are now specialised per target language, terminology is deduplicated per entity across a video, and a guard catches drift between languages that share a script (English against Spanish, French, German, Portuguese, Italian).
- Speaker labels no longer leak into translated text, and speaker identity no longer changes across window boundaries.
- Background music is more stable. Loudness is now locked per task instead of being continuously normalised, which removes the pumping that could be heard under dialogue.
- Browser extension: a live countdown to first playback, a notification when the dub is ready, and links that jump between the extension card and the web workspace.
- Fixed pricing and billing wording that described charging by source video duration. Billing has always settled on delivered audio; the copy was wrong, not the behaviour.
June 2026
- Per-speaker voice reference lock: one reference per speaker within a transcription call, which removes the timbre jumps that could happen mid-video.
- Reference audio is now chosen by perceptual quality (DNSMOS P.835 signal rating) plus acoustic admission and speaking-rate checks, rather than by position alone.
- Subtitle editing gained a manual per-segment re-dub that uses that segment's own original audio as the voice reference.
- Background music separation moved to a higher-quality model.
- Faster dubbing across the board: continuous-batching TTS, batch-level pipelining, and warm-up coverage for the first batch.
- Transcription defaults switched to Scribe, with in-window pipelining so translation starts before the whole window is transcribed.
May 2026
- Live stream dubbing with a rolling DVR window.
- Subtitle editing rebuilds only the lines you changed, rather than the whole video.
- MP4 export with soft-baked packaging, and parallelised container IO on the export path.
Something missing or broken?
This list only covers changes you can notice as a user. If something regressed, tell us: support@waveshift.net
See also: About WaveShift.