Quick Translate workflow

Quick Translate runs two services back to back:

1. Transcription — speech recognition produces a transcript with segment timings.
2. Translation — those segments are translated as continuous discourse and redistributed across the same cues.

This is the difference that matters, and the reason it costs less than Subtitling.

Skipping forced alignment means cue timings come from speech recognition rather than from the aligner. They are measured against your audio, not invented — but they are coarser, and you get fewer, longer cues. The same ten-minute recording produced 171 cues through Subtitling and 28 through Quick Translate.

Cue start and end times are copied verbatim from the source and are never taken from model output. Word-level timings are not emitted: translated words have no audio of their own, so any per-word timestamp for them would be fabricated.

Set autoApprove: false to pause after transcription and correct the transcript before it is translated — worth doing when names or jargon matter, since a recognition error becomes a translation error. A paused workflow waits indefinitely and is never auto-refunded.

One upfront charge of £0.06 per minute of source audio, with a one-minute minimum and per-minute rounding. Register and reading-speed limits are included, not upgrades. If a stage fails terminally, what completed is charged at standard rates and the rest is refunded.

Source languages are the aligner-supported set, the same as Subtitling. Targets are the wider translation set — see Translation.