Methodology

How Subtitle Translator works

ASR vs LLM roles, hosting, deletion, limits, and what we never promise.

Pipeline

  • Transcription (media tab only): speech-to-text with timestamps via a configurable ASR backend (local Whisper or a hosted ASR service). Used for transcription only — not for target-language translation.
  • Translation (both tabs): translation via hosted LLM or third-party machine translation services (provider selected by server config). Cue numbers and timestamps are returned unchanged; only cue text is translated.

Billing

Billed per minute, rounded up, minimum 1. Media jobs bill by audio duration. SRT/VTT jobs bill by the end time of the last cue. Transcription + translation is a single charge (not doubled).

Files & privacy

Files are uploaded to our servers, processed by the configured ASR/MT providers, then deleted within 1 hour after the job completes or fails. Not used for training. Not human-reviewed.

Limits

  • Max 60 minutes per job
  • Media max 200 MB; SRT/VTT max 2 MB
  • Chinese / Japanese / Korean options appear as “Coming soon” until ENABLE_CJK is turned on after quality probes
No-promise list: we do not promise accuracy percentages, platform approval (YouTube or otherwise), “better than human” quality, or that translation will be publication-ready without proofreading.