Subtitles

flowchart LR
  A[Audio or video] --> B[Transcribe]
  B --> D[Review written text]
  C[Existing subtitles] --> D
  D --> E[Translate if needed]
  E --> F[Check text and timing]
  F --> G[Export SRT or VTT]

Transcription and timing

Use Quick Transcribe for a direct transcription, or a subtitle project when you need correction, translation and later revisions. Select the source language explicitly when you know it.

Whisper, Parakeet, Qwen3-ASR and MOSS have different language and timing support. A recognizer can understand a language without its aligner supporting subtitle timestamps. Qwen3-ASR has 30 recognition languages in this release; 20 have a timing route and ten are transcript-only. Check the catalogue before choosing it for subtitles.

Review the written version

Read names, dates, abbreviations and numbers against the source. Correct the source transcript before translating when practical. Review target-language phrasing and caption duration, then select the version for export.

Keep ordinary spelling and digits in captions. Pronunciation respellings and expanded spoken numbers belong to the speech layer. Shortened captions should preserve the meaning of the spoken line.

Export

SRT is widely supported. VTT works with web players. A video can include selectable subtitles or burn them into the image. Select the source, translation or dual-language output you need, and check line wrapping in the exported file.

A practical review pass

1. Check the source language and the selected transcript. An imported SRT already has cue timings; a plain transcript still needs a timing route.
2. Play a short section while reading the editable revision. Check omitted words and speaker changes before polishing punctuation.
3. Correct the source, then translate if needed. The source and translation remain separate versions.
4. Review long cues, fast speech, overlaps and the final cue. A correct sentence can still be uncomfortable to read at its assigned duration.
5. Export and open the actual SRT or VTT in the intended player or editor. Check encoding, line breaks and synchronization.

Speakers and voices

Speaker labels describe who spoke in the recording. They can help organize a transcript or a later voiceover, but a subtitle-only project does not need a synthetic voice. A speaker label is also not proof of a person's identity.

When words and timing disagree

Choose an engine with a timing route for the source language. Recognition, speaker diarization and word alignment are different stages. For MOSS, the model card claims more than fifty languages without publishing a full list; native speaker-attributed timings and subsequent word alignment have different limits. Check a representative clip before a long transcription.

See the subtitle tutorial.