✳ FREE · TO ENGLISH · NO UPLOAD

Translate audio or video to English, free in your browser

Whisper can listen in one language and write English. Drop a file: the translation runs on your device and you get English text and subtitles with timestamps.

Public-domain LibriVox recording. Load it, then press Transcribe to try the full tool.

Translation uses Whisper's built-in speech-to-English mode. With this small model it gives the gist, not a publishable translation, and it only translates into English.

First use downloads the speech model and engine (about 97 MB, of which 72.5 MB is the model); later visits load it from your browser cache. Runs on your CPU; nothing is uploaded. Up to 100 MB and 2 hours per file; under 30 minutes is the comfortable range on a laptop.

To translate speech to English for free, drop a file above with “Translate the speech to English” selected: Whisper’s built-in translation runs in your browser and writes English text and subtitles. With this small model, treat it as a gist. It only translates into English, and for publishable subtitles transcribing first and translating the text works better.

Sources: Whisper paper (Radford et al., 2022), Appendix D (checked 2026-09-25) · openai/whisper README (checked 2026-09-25)

Checked by Koldflux · updated 2026-09-25 · BLEU scores read from the Whisper paper, Table 14

Translation quality into English, by spoken language

BLEU score of Whisper’s speech-to-English translation on FLEURS (0 to 100, higher is better). The last column is the largest model the paper measures, for comparison. Below about 5, treat the output as unusable.

Spoken languageWhisper base (this tool)Whisper large-v2
Spanish12.423.3
French13.132.2
German13.734.6
Portuguese22.438.1
Italian11.323.6
Russian12.127.8
Dutch7.524
Polish5.122.3
Turkish2.426.6
Japanese1.518.9
Chinese1.918.4
Korean3.721.3
Arabic125.5
Hindi3.422
Tamil0.49.2
Vietnamese0.920.4
Indonesian2.629.1
Ukrainian6.929.4

What the translation looks like

On the first 30 seconds of a public-domain Spanish LibriVox recording, the Spanish transcript was nearly perfect, while the direct English translation read: “The Soap Fables recorded for the book box.org by Karen Savage. Fable #9 The Sour and the Spino…”. The original says “Las fábulas de Esopo, grabado para LibriVox.org por Karen Savage. Fábula número 9. La zorra y el espino” (Aesop’s fables, recorded for LibriVox.org; fable 9, the fox and the hawthorn). On a German poem the first pass got stuck repeating “the spirit of the spirit”, and the tool’s fallback decoding produced a readable but loose sentence instead.

That is the pattern to expect: the transcript in the original language is much better than the direct translation. For anything you publish, transcribe first, then translate the text.

Two ways to get English

  1. Quick gistChoose “Translate the speech to English” and run. Good for understanding what a clip is about.
  2. Publishable subtitlesTranscribe in the original language, correct it, download the SRT, translate the text lines with a text translator, and keep the timestamps.
  3. Check the timingEnglish is often shorter or longer than the original; run the subtitle checker for cues that are now too fast.

Limits and when to use something else

Frequently asked questions

Can it translate into languages other than English?

No. Whisper’s built-in translation only produces English. To go from English to another language, transcribe in English and translate the text.

How good is the translation?

For this small model, rough. The Whisper paper scores Whisper base at BLEU 12 to 14 for Spanish, French and German into English, and below 4 for Japanese, Chinese, Arabic and Hindi. The largest model in the paper roughly doubles the European scores.

What is the better route for subtitles?

Transcribe in the original language here, correct the transcript, then translate the text with a text translator and put it back into the SRT, keeping the timestamps.

Does Koldflux Studio translate?

No. Studio transcribes with Whisper large-v3-turbo, which OpenAI says is not trained for translation, and it does not offer translation or dubbing.

Is anything uploaded?

No. Translation runs in your browser tab like the transcription.

Language pages

Each page has the tool preset to that language, measured accuracy, our own test on a real recording and subtitle rules.

FROM RECORDING TO POSTS

Turn one idea into a week of posts

Koldflux Studio (paid, from $19/month) transcribes uploads in their original language on our servers with Whisper large-v3-turbo, then turns one recording into Shorts and carousel posts you review, approve and export. It does not translate, publish or schedule.

  1. 1 · Recording, idea or script
  2. 2 · Pick the angles
  3. 3 · Review and approve
  4. 4 · Export Shorts and carousels
See Koldflux Studio
Koldflux Studio: choosing which angles (Opportunities) to turn into Shorts before anything is generated
Koldflux Studio: you choose the angles before anything is made.