Translate Audio

Turn speech into a voice-over in another language, timed like the original.

Processed in your browser — files never leave your device

Drop a file here

or

Select file

About Translate Audio

Upload a recording or video and get it back spoken in Hindi, English, Bengali, Marathi, Tamil, Telugu, Urdu, Spanish, German and 10 more languages, in a male or female AI voice. Each sentence is placed where it was said, and the voice follows the speaker's pitch. Free, in your browser — nothing is uploaded.

How Translate Audio works

Translate Audio works in three steps, all in your browser. First Whisper, OpenAI's open-source speech model, listens to the recording and writes what is said in English, sentence by sentence, with the time each sentence starts. It understands about 100 spoken languages, including Hindi, Bengali, Tamil, Telugu, Marathi, Urdu, Spanish and Arabic.

Next an open-source Marian translation model (OPUS-MT, from the University of Helsinki) translates the sentences into the language you choose, and Kokoro, an open-source text-to-speech model, reads each one aloud in a male or female voice. Kokoro has its own voices for six languages; for the others the sounds of the language are read by the closest voice, which gives an accent. For English the translation step is skipped.

Finally each spoken sentence is placed at the moment the original sentence was said. If a translation is longer than the gap it has to fit, it is sped up slightly without changing the pitch. With Match the speaker, a male or female voice is chosen from the original speaker's pitch and nudged towards it, so a deep voice stays deep. For a video, the picture is kept exactly as it is and only the sound is replaced.

The models are downloaded once from the Hugging Face CDN (about 450 MB in total the first time, less for later languages) and kept by your browser. Your file is never uploaded.

When to use Translate Audio

  • Turn an English lecture, tutorial or product video into Hindi for your audience — or a Hindi one into English.
  • Make a voice-over of a reel or short in another language to reach new viewers.
  • Understand a voice note, interview or podcast clip in a language you don't speak.
  • Get translated subtitles (SRT) and a written translation alongside the new audio.

How to use Translate Audio

  1. Add your audio or video. Drop an MP3, WAV, M4A, MP4 or other file of up to 15 minutes into the box above.
  2. Choose the language. Pick the language to translate into. Leave the spoken language on Detect automatically, or choose it.
  3. Translate and download. Click Translate, listen to the result and download it — plus the translated text or subtitles if you want them.

Options explained

Translate into
With a natural voice: English, Hindi, Spanish, French, Italian and Portuguese. With an accented AI voice (beta): Bengali, Marathi, Gujarati, Punjabi, Tamil, Telugu, Kannada, Malayalam, Urdu, German, Russian and Indonesian.
Spoken language
Detect automatically works for most files; choose the language yourself for very short clips or mixed speech.
Voice
Match the speaker picks a male or female voice from the original speaker's pitch and moves it a little towards theirs. Or choose Female or Male yourself (French has a female voice only).
Original sound
Remove replaces the sound completely. Keep quietly underneath leaves the original audio at a low level under the new voice, like a news interpreter.

Frequently asked questions

Will the new audio sound like my own voice?

No. The voice-over is read by an AI narrator; it follows the speaker's timing and, with Match the speaker, their gender and pitch, but it does not copy their voice. Copying a real person's voice needs much larger paid models.

Which languages can I translate into?

Eighteen: English, Hindi, Spanish, French, Italian and Portuguese have their own natural voices; Bengali, Marathi, Gujarati, Punjabi, Tamil, Telugu, Kannada, Malayalam, Urdu, German, Russian and Indonesian are read by the closest AI voice, so they have an accent — check the result, and use the subtitles it also makes if needed. The original can be in any of about 100 spoken languages.

How good is the translation?

Clear speech gives understandable, natural results for everyday talk, lessons and explanations. Names, slang, jokes and technical terms can come out wrong, so listen to the result or read the translated text before you publish it.

Does it work on videos?

Yes. Upload a video and you get the same video back with the new voice; the picture is copied without re-encoding when the file is MP4, MOV or MKV.

How long can the file be?

Up to 15 minutes. Translation runs on your device, so a computer is much faster than a phone; a 5-minute clip takes a few minutes on a typical laptop.

Is my file uploaded?

No. Listening, translating and speaking all run in your browser. Only the language models are downloaded, once, from the Hugging Face CDN.

Can I get the translation as text or subtitles?

Yes. After translating, download the translated text or an SRT subtitle file timed to the original speech.

Is Translate Audio free?

Yes. Translate Audio on DocEpic is free, with no sign-up, no watermark and no limit on how often you use it. The site is supported by ads.

Are my files safe with Translate Audio?

Yes. Translate Audio runs entirely in your web browser, so your files never leave your device and are never uploaded to us.

Do I need to install any software?

No. Translate Audio works in any modern browser on Windows, macOS, Linux, Android and iPhone.

Is there a file size limit?

Files up to 1024 MB are accepted. Because the work happens on your device, very large files depend on your device's memory.

Related tools