How to translate subtitles, step by step

This is the practical version: which button, in which order, and what each screen will show you. It takes about five minutes to read and covers the whole path from upload to two subtitle files that line up.

Before you start

Two things decide whether you need any of this.

If all you want is English out of non-English audio, stop here. You do not need the translation feature. In the transcribe dialog, tick Transcribe to English — it runs inside the speech model itself and is available on every plan, including the free one. There's a dedicated page for that route: translate audio to English.

Everything else is the Max plan. Translating a finished transcript into one of 56 target languages — including English alongside the original — is a Max feature, and the pricing page spells out what each plan includes. The rest of this guide assumes you're on it.

Step 1 — Choose the target language before you submit

Open Transcribe file, pick your audio or video, then expand Speaker detection & more settings. Inside is a dropdown labelled Translate transcript to, which starts at No translation. Choose your language there.

This is the step people get wrong, so it's worth being blunt about it:

The language is fixed once the job starts. You cannot add a second language later, and you cannot switch it. There is no "translate this now" button on a finished transcript.

It's stricter than it sounds, because the file you uploaded is cleaned up shortly after the job succeeds. So changing your mind doesn't mean clicking a different option — it means uploading the file again and re-transcribing it from scratch, spending the minutes a second time. Decide before you press Transcribe.

While you're in that panel: if there is more than one voice in the recording, tick Detect speakers too. Subtitles then break where the speaker changes instead of mid-answer, which makes the translated track much easier to read.

Step 2 — Let it run

Transcription happens first, then translation. You'll see Translating into {language}… on the job while the second half runs, and the job only counts as finished once both are done.

If translation fails, the transcript survives. The page says so explicitly — "Translation failed. The transcript below is unaffected." — and you still have a complete, timestamped transcript to export.

Step 3 — Proofread the original, and only the original

Open the transcript. Click any line to play the audio from that point, which is the fast way to check a name or a number you're unsure about.

Two rules here:

Switch back to Original before you edit anything. The editable document is the source-language transcript. The translated view is generated output.

Editing the original does not re-translate it. Fix a name after the translation is done and the page will warn you: "The transcript was edited, so this translation may be out of date." That's a flag, not a repair — the translation stays as it was. If a proper noun really matters, it's cheaper to fix the translated line later in your subtitle editor than to re-run the whole job.

Step 4 — Switch between the two versions

Once the translation is ready, a row of buttons appears above the transcript: Original, then one per language.

Clicking a language swaps the text and nothing else. Timestamps and speaker labels come from the same one-to-one segment list, so click-to-play keeps working exactly the same way in the translated view. That's the whole design: one set of cue times, two sets of words.

Step 5 — Export both tracks

Here's the mechanic that isn't obvious: export gives you whichever version is currently on screen.

So getting a bilingual pair is two exports, not one:

1. Stay on Original → export SRT.
2. Click your language → export SRT again.

You now have two files whose cue times match line for line. That's what makes them stackable: in Premiere, Final Cut or DaVinci put them on two separate subtitle tracks, one above the other. On YouTube, upload them as two subtitle tracks and let the viewer pick — the same pair of files works for translating a YouTube video.

If you'd rather start from an existing subtitle file than from the video, the SRT translator page covers that path.

The same applies to all eight formats. If you want the translation as a document rather than subtitles, switch to the language first, then export PDF or DOCX.

Three things that will bite you

The language choice is permanent. Worth repeating, because it is the only mistake in this workflow that costs you a full re-run.

Line lengths change. A line that fits comfortably in English can overflow in German or Spanish, while Chinese and Japanese usually come out shorter. If your platform enforces a character limit, fix it in your subtitle editor after export — the in-app editor won't help you here, since it edits the original.

Transcription is not the gated part. Speaker recognition, audio restoration, link import and all eight export formats work on every plan, free included. Only the 56-language translation needs Max.

Ready to try it? Start from subtitle translation if you're working from video, or translate video to text if you want the transcript as a document rather than as cues.