SUBTITLE GENERATION
How to generate subtitles in different languages
Spoken language is the language in the recording. It is not the language you want the subtitles translated into. Pick it before you generate. Translate, on Pro or Max, is a later step that rewrites cue text without listening to the audio again.
The video is already in English and you want Spanish subtitles? Generate in English first. Translation is a Pro or Max action on the finished cues. See Translate on Pro
Do it in this order
- 01
Decide which job you are doing
If people are speaking Spanish, set Spoken language to Español and generate. If people are speaking English and the captions should read Spanish, generate in English, review the cues, then use Translate. Mixing those two steps is how a transcript ends up in the wrong language.
- 02
Pick from the spoken-language list
The control includes English, 中文, Español, Português, Français, Deutsch, 日本語, 한국어, हिन्दी, Hinglish, Italiano, Nederlands, Русский, العربية, Türkçe, Tiếng Việt, ไทย, Bahasa Indonesia, Polski, Українська, and Svenska. The choice is remembered on this device. Default is English.
- 03
Generate once in that language
A wrong choice still spends the processing minutes and returns a busy, incorrect transcript. If the script is obviously wrong, change Spoken language and run transcription again. Do not try to repair a Chinese video that was transcribed as English by editing every line.
- 04
Use Translate only after the source cues are right
Translate sends the subtitle text, not the video, to a translation pass and replaces the cue text. Timing stays. It requires Pro or Max and spends processing minutes once per run. High Accuracy is separate: it uses a stronger speech model at 2× the minutes, and you turn it on before you transcribe.
What usually goes wrong
Leaving the default on English for a non-English video
English is the default because this is the overseas site, not because the model detects the language. A short Chinese clip can come back empty or in the wrong script until you select 中文.
Using Translate as a substitute for the spoken language
Translate does not re-listen. If the source cues are wrong, the translation is a fluent version of the wrong words.
Expecting one file to carry two spoken languages
Hinglish is the mixed option for Hindi plus English. Other code-switching has to be generated in the language that should dominate the transcript.
Questions this page should settle
Is there an automatic language detector?
No. You set Spoken language before generation. If the result is in the wrong language, change the setting and run again.
Where are Hindi and Arabic handled as their own pages?
The Hindi and Arabic subtitle tools use the same editor. They are starting points for those spoken languages, not a different transcription engine.