Skip to content
Translation

How to translate a video into English subtitles

· 5 min read · Total Soft Solutions

A video in Spanish, Japanese or Russian can reach a much bigger audience with English subtitles, and a video you cannot understand becomes watchable the moment the words appear. The usual route is slow: transcribe the speech, paste the text into a translator, then rebuild the timing by hand so it lines up with the picture.

There is a shorter path. A multilingual speech model can listen to speech in one language and write English directly, with each line timed to the moment it was spoken. This guide shows how to use that to make an English subtitle file, how to check the result, and where the method falls short. Everything runs in your browser, so the video is never uploaded.

How speech-to-English translation works

Most translation tools work in two steps: first transcribe the speech, then translate the text. The multilingual models in this tool can do both in one pass. They listen to the speech and produce English text straight away, and the timing of each line is kept from the audio.

That has a practical advantage for subtitles. Because the English lines are timed by the speech itself, you can export them as subtitles without matching a translated script to the video by hand. It also has a limit worth knowing: the translation goes only to English. For other languages there is a separate route, covered below.

What you need

  • A video or audio file. Common formats such as MP4, MOV, MKV, WEBM, MP3 and WAV work, and only the sound track is used.
  • A multilingual model. The English-only models cannot translate. The Models screen shows which are multilingual, and the guide to Fast, Standard and High accuracy explains the sizes.
  • A model of Standard size or larger for most jobs. Fast often gives loose or partial translations.
  • A language OmniaScribe understands, which covers 99 of them, though quality varies by language.

Try it on your own file

or drop it here · MP3 WAV M4A MP4 MKV MOV FLAC OGG
Nothing is uploaded

Step by step

  1. Open the translate audio to English page and pick your file. The Translate to English option is already switched on for this one file.
  2. Choose the spoken language yourself instead of using detection when you can, particularly if the video begins with music, silence or a greeting in a different language.
  3. Check that your selected model is multilingual. If it is English-only, the workspace tells you so and you can pick a different model on the Models screen.
  4. Start the job. The English lines appear as it works, timed to the speech.
  5. Read the result against the video, fix anything wrong with Edit, then export SRT or VTT. The guide to SRT and VTT helps you choose the file type.

The subtitle generator starts transcribing as soon as you pick a file, and it does not translate on its own. To use it for English subtitles, first open the workspace, switch on Translate to English under Before you start, then pick the file there. The result is the same.

Checking the English

The output is a faithful working translation, not a polished one. For clear speech the meaning usually comes through well. But machine translation has predictable weak spots, and viewers will notice them in subtitles.

  • Idioms are often translated word for word, which can read as nonsense.
  • Names of people and places may be spelled the English way, or oddly.
  • Very short lines can lose context, because there is less to go on.
  • Numbers, dates and units can be wrong. Check every one against the audio.
  • Formal and informal address, which some languages mark and English does not, can disappear.

A good habit is to run the file twice: once as a plain transcript in the original language, and once translated. Reading them side by side shows where a line was translated loosely, and a speaker of the language can check the original quickly because it is already written down and timed.

Getting a better translation

The same things that help a transcript help a translation, and they matter more, since an error in the heard word becomes an error in the English. Prefer a clean audio track, a close microphone and a model one size up from your first try. A video with background music under the voices is one of the hardest cases.

Recordings that switch between languages part-way through are difficult to translate cleanly. If you can, split the file at the point of the change and process each part with the right language. If the file is long, it may exceed what your model handles on your device. The workspace warns you before it starts, and splitting at a natural break is the fix.

Finally, shorten subtitles that go by too fast. English is often shorter or longer than the original speech, so a cue can end up too dense to read. Trim the wording in Edit so each cue can be read comfortably in the time it is on screen.

Subtitles in languages other than English

The speech model translates into English only. If you need subtitles in another language, transcribe the video in its original language first. Then use Translate on the result. In Chrome on a computer, the browser’s own on-device translator makes a translated copy that keeps the same timestamps, and you can export it as SRT or VTT too.

OmniaScribe never sends the text to an online translation service. The trade-off is that this route depends on your browser offering a translator for the language pair, so it is not available everywhere. Check the result with a native speaker, as you would for any translation.

Putting the subtitles on your video

Once the English file is checked, attach it where your video lives. Most platforms have a subtitles section in the video’s settings where you upload an SRT or VTT file and label it English. A video editor can import the same file as a captions track. Keep the original-language transcript too, so you can fix a line later and export again.

If the video came from a longer source, such as a video to text job on the whole recording, remember that subtitles must match the exact version of the video you publish. Trim first, then transcribe, or the timings will drift.