Skip to content

Audio to Text: Transcribe Audio and Video

Transcribe a recording or video with Whisper AI in your browser: Hindi, English and 90+ languages. Copy the text or save TXT, SRT or VTT subtitles.

Private · on device. Runs in your browser: your file is not uploaded. No sign-up.

Audio or video: MP3, M4A, WAV, OGG, WebM, MP4 and more

Speech to text, on your own device

This tool uses Whisper, the open speech recognition model released by OpenAI under the Apache-2.0 licence, running inside your browser. Most transcription websites upload your recording to their servers; here the model comes to you instead. It is downloaded once from Hugging Face, kept by your browser, and your recording is processed locally.

What files work

Anything your browser can play: MP3, M4A, AAC, WAV, OGG, Opus, FLAC and WebM audio, and the sound track of MP4, MOV and WebM videos, including WhatsApp voice notes and phone recordings. The sound is converted to the 16 kHz mono format Whisper expects. Recordings of up to three hours work on a computer; on a phone, shorter is safer.

Languages

Whisper recognises about 99 languages and detects which one is spoken. Choosing the language yourself is faster and avoids mistakes when a recording starts with music or silence. It can also translate speech into English text as it goes; translation quality varies by language.

  • Indian languages: Hindi, Bengali, Marathi, Tamil, Telugu, Gujarati, Kannada, Malayalam, Punjabi, Urdu, Nepali and Assamese, written in their own scripts. Use the Small model for the best results.
  • Mixed Hindi and English (Hinglish) is common in Indian recordings. Whisper usually writes it in Devanagari with English words in between; choose English instead if you want it in Latin letters.

Downloads: TXT, SRT and VTT

  • TXT is the plain transcript in paragraphs, optionally with a time at the start of each line.
  • SRT is the subtitle format read by almost every video editor, player and YouTube.
  • VTT (WebVTT) is the subtitle format for video players on web pages.

Need the reverse? Text to speech reads English text aloud. For text in a photo or scan, use image to text.

Frequently asked questions

Is my recording uploaded anywhere?

No. The Whisper speech recognition model is downloaded to your browser once, and your recording is transcribed on your own device. Interviews, meetings and voice notes stay private.

Does it work for Hindi and other Indian languages?

Yes: Whisper understands Hindi, Bengali, Marathi, Tamil, Telugu, Gujarati, Kannada, Malayalam, Punjabi, Urdu and more, writing each in its own script. Accuracy for Indian languages is clearly better with the Small model; Tiny is fine for clear English but makes many mistakes in other languages. Odia is not one of Whisper's languages.

Which model should I use?

On a phone, Tiny (about 44 MB). On a computer, Base (about 80 MB) is a good balance, and Small (about 250 MB) is the most accurate, especially for Indian languages and accented English. Where the browser can use the graphics chip (WebGPU), larger and much faster versions are downloaded instead (about 120, 210 and 590 MB respectively). The exact size is shown before anything is downloaded, and each model is downloaded once and kept by your browser.

How long does it take?

On a recent computer, Tiny and Base usually transcribe faster than the recording plays; Small is slower. A phone may take longer than the recording itself. The progress bar shows how far it has got, and the transcript fills in as it goes.

Can I make subtitles for a video?

Yes. Choose the video file itself: the tool reads its sound track. Download SRT for video editors, YouTube and most players, or VTT for web players. Long sentences are split into caption-sized lines with their own times.

How accurate is it?

Clear speech with little background noise comes out well; crosstalk, music, heavy accents and poor microphones cause mistakes, and names are often misspelt. Always read the transcript through before you rely on it.