Version 3.0.0
Dubbing, Workflows, two new speech engines and live transcription from your microphone.
Dubbing (beta) Windows
Give a video a new spoken language, entirely on your computer. The app separates the dialogue from the music and effects, works out who is speaking, translates each line to fit the time it has, and speaks it in a voice copied from that speaker. The new dialogue is mixed back over the original music and effects and added to the video as an extra audio track — the original stays untouched. Speaker names in SDH subtitles are used to tell voices apart, sound descriptions like [door slams] are never spoken, and a voice library keeps voices across projects so you can back them up and share them. Runs best on an NVIDIA graphics card.
Workflows
Chain the app's tools into one pipeline and run it on a file or a whole folder: transcribe, extract or OCR a track, clean up, translate, shift, recolour, reposition, sync, embed and export. Start from a template — several languages at once, a full subtitle package, OCR then translate — or build your own. Italics and ASS styling survive every step, retrying a run redoes only the files that did not finish, and the results can be exported as JSON, CSV or text.
Two new speech engines: Qwen3-ASR and Parakeet Windows
Alongside Whisper you can now transcribe with Qwen3-ASR, strong on Chinese, its dialects and accented English, or NVIDIA Parakeet, a fast and accurate engine for English and 24 other European languages. Both detect the spoken language themselves and run on the CPU; long recordings are split across several processes to finish sooner. Choose and install them on the new Speech Engines screen, or straight from the model list on Transcribe.
Live transcription from your microphone
Transcribe has a second mode that writes subtitles as you speak, phrase by phrase, with Whisper. The app recommends a model your computer can keep up with, and the transcript can be saved as SRT, WebVTT, ASS or plain text.
A dashboard of your work
The app opens on a dashboard: subtitles created and translated, hours of speech transcribed, a day-by-day chart, and the languages and models you use most. Only counts are kept, on your computer — never file names or subtitle text.
Also in this version
- The main menu is grouped by what you want to do: get subtitles, edit and improve, output, and automate.
- Embedding subtitles into a video that carries attached fonts no longer fails with "Invalid argument".
- Live transcription cuts long speech at its quietest moment instead of mid-word.
- Parakeet warns you when it is given a language it does not recognise, rather than writing nonsense.
- Qwen3-ASR no longer uses the vocabulary hint, which could make it replace a sentence with the list of names.