Subtitle translator

Translate an SRT or VTT file and keep every timestamp. Compare each line with the original and fix it before you download.

También en español

Cloud translationTiming stays exactly as in your file.
Glossary (optional)

Your captions are sent to SoundHalo and translated by the DeepSeek model through OpenRouter. OpenRouter only routes them to hosts it lists as not storing or training on prompts. SoundHalo does not save the captions or the translation. Privacy policy

Example: English to Spanish

Four captions from a microphone review, translated to Latin American Spanish by this tool. The times did not change, and the product names, numbers, units, app name and italic tag came through as written.

TimeOriginalTranslation
00:00:03,400 --> 00:00:06,900Today we're comparing three USB microphones under $150: the Rode NT-USB Mini,Hoy comparamos tres micrófonos USB por menos de $150: el Rode NT-USB Mini,
00:00:10,200 --> 00:00:13,600Each one was recorded at 48 kHz, 24-bit, about 15 centimetres from my mouth.Cada uno se grabó a 48 kHz, 24 bits, a unos 15 centímetros de mi boca.
00:00:24,200 --> 00:00:27,500<i>[keyboard clicking]</i><i>[teclado haciendo clic]</i>
00:00:41,900 --> 00:00:45,200One tip: turn off "auto level" in the MOTIV app before you record a podcast.Un consejo: desactiva "auto level" en la app MOTIV antes de grabar un podcast.

What stays the same

The number of captions, their order and every start and end time are guaranteed. The tool checks each reply and translates a group again if any caption is missing, merged or out of order. Times are copied from your file, never from the model.

The model is also told to keep line breaks, formatting tags such as <i> and <b>, and names, numbers, units and web addresses as written, unless your glossary says otherwise. It usually does, but this is not checked, so review those lines. In a Japanese test, for example, it wrote centimetres in Japanese and moved a line break.

Captions are translated in groups of up to 40, with the two captions before each group as context, so sentences that run across captions still read naturally. A sentence split across two captions stays split across the same two captions.

What to check before you download

The editor shows the original and the translation side by side, and flags lines worth a second look:

  • Empty translations. These block the download until you fill them in.
  • Lines that came back unchanged, which can mean a missed translation.
  • Lines faster than 25 characters per second. Translations are often longer than the original, so a caption that was comfortable to read can become too fast. Shorten the wording rather than changing the times.

Also read for tone: formal or informal "you", gendered words, jokes and references to things on screen. After downloading, you can run the file through the subtitle validator to check overlaps and reading speed again.

Keep names and terms consistent

Open Glossary to list terms, one per line. Write a name on its own to keep it unchanged, such as SoundHalo. Write source = translation to fix how a term is translated, such as mic = micrófono. The glossary is sent with every group, so the term is handled the same way through the whole file.

Starting from audio or video?

Create captions first with the SRT generator, correct them, then choose Translate captions. The captions open here in a new tab, ready to translate. Translating a corrected transcript gives a better result than translating raw speech recognition.

Free to try, more with an account

  • No account: up to 6,000 characters per file, about six minutes of speech, and 5 files a day. This allowance is shared by your network and by all visitors, so it can run out.
  • Free account: 60,000 characters per file, about an hour of speech, and 20 files a day.
  • Pro: 150,000 characters per file, enough for most feature films, and 200 files a day.

Related tools

FAQ

Does the translation keep my timestamps? +

Yes. Every caption keeps its number, start time and end time. Only the text changes. The tool sends each caption with a fixed id and checks that the same ids come back in the same order. If a reply drops, merges or reorders captions, that part is translated again. The times in your download are copied from your file, not from the model.

Which subtitle formats can I translate? +

SRT and WebVTT (.vtt). You can download the result as either format. Files must be smaller than 2 MB. ASS, SSA, SUB and SBV files are not supported, so convert them to SRT in your subtitle editor first.

How good is the translation? +

It is machine translation by an AI model, so treat it as a strong draft. It handles everyday speech well. It is told to keep names, numbers and tags unchanged, but check them. Check jokes, idioms, slang, formal and informal address, and lines that depend on what is on screen. Every line is editable before download.

What are the limits? +

Without an account: up to 6,000 characters per file (about six minutes of speech) and 5 files a day per network. This free allowance is shared by everyone on your network and by all visitors, so it can run out for the day. Free account: 60,000 characters per file and 20 files a day. Pro: 150,000 characters per file and 200 files a day. Limits reset at midnight UTC.

Where are my subtitles processed? +

The caption text is sent to SoundHalo and translated by the DeepSeek V4.1 Flash model through OpenRouter. SoundHalo asks OpenRouter to use only model hosts that do not store or train on prompts. SoundHalo does not save your captions or the translation. Do not upload captions you are not allowed to share with a third-party service.

Can I translate subtitles from a YouTube link or burned into a video? +

No. The tool needs a subtitle file. If a video has no subtitle file, create one from its audio with the SRT generator first, then use Translate captions in the editor.

Can I translate into several languages at once? +

One language at a time. After downloading, choose Translate to another language. Your original file stays loaded.