Transcribe Audio and Video to Text
Get a transcript of an interview, lecture, meeting or voice note, with timestamps and subtitle files. Speech recognition runs in your browser, so your recordings stay on your device.
- Downloads the speech model from Hugging Face once; your audio and video never leave the device
- Free, no sign-up
- Works on phones
You’re offline. This tool needs an internet connection for some features. They will work again when you reconnect.
Looking for a speech model on this device…
What goes online
- Downloads the speech model from Hugging Face once; your audio and video never leave the device.
- Everything else happens inside this browser tab.
Often used next
Looking for something else?
DeskTray has hundreds of free tools that work the same way.
How to transcribe audio or video to text
- Download a speech model once in Settings (the tool takes you there). Whisper Tiny is about 42 MB; bigger models are more accurate.
- Choose an audio or video file, such as an MP3, a voice note or an MP4.
- Pick the spoken language, or leave it on “Detect automatically”. Choose “English translation” to get English text from another language.
- Press Transcribe. The text appears part by part, with the time each line starts.
- Copy the text, or download it as a TXT file or as SRT or VTT subtitles.
Features
- Runs on your device: the audio and video never leave it
- Hindi, Bengali, Telugu, Marathi, Tamil, Urdu, Gujarati, Kannada, Malayalam, Punjabi and about 90 more languages
- Translate speech in another language to English text
- Timestamps you can tap to play from that moment
- TXT, SRT and VTT downloads
- Long recordings are done 30 seconds at a time, with progress and a cancel button
How it works, and what is downloaded
This tool uses Whisper, the open speech recognition model from OpenAI, running inside your browser. The first time, you download a model in Settings (the tool takes you there); it comes from Hugging Face (the onnx-community builds) and is kept in your browser’s storage, so next time it starts straight away, even offline. Settings also deletes a model to free the space.
Your recording itself is never sent anywhere. The sound is decoded on your device, cut into parts of up to 30 seconds at natural pauses, and each part is turned into text with the time it starts. Silent parts are skipped. Where the browser offers WebGPU (recent Chrome or Edge, on computers and many Android phones), the model runs on the graphics chip, which is much faster; elsewhere it runs on the processor.
Which model to choose
- Whisper Tiny: 42 MB (117 MB with WebGPU). The quickest. Fine for clear English, but often wrong in Hindi and other Indian languages.
- Whisper Base: 76 MB (199 MB with WebGPU). Noticeably better English. Indian languages are still weak.
- Whisper Small: 240 MB (288 MB with WebGPU). Good English, and Hindi and Urdu come out mostly right when the speech is clear. Tamil, Telugu, Bengali, Malayalam and the others still have many mistakes.
- Whisper Large v3 Turbo: 727 MB, WebGPU only. The most accurate and the best choice for Indian languages, though not perfect. It transcribes only: it was not trained to translate.
Speed depends on the device. In our tests on a computer’s processor, without WebGPU, one minute of speech took about 7 seconds with Tiny, 15 seconds with Base and 40 seconds with Small. Phones are slower; a graphics chip is much faster. Long files are fine: the progress bar shows how much is done and roughly how long is left, and you can cancel at any time.
How accurate is it?
Clear speech close to the microphone in English gives very good results with Base or Small. Accuracy drops with background noise or music, several people talking at once, strong accents, and phone-call quality audio. Names, numbers and technical words are the usual mistakes, and mixed Hindi and English is often written in just one script.
Whisper sometimes invents a sentence where there is no speech, such as “Thank you.” at the end of a recording; silent parts are skipped to avoid most of this. Always read the transcript before you rely on it, especially for anything legal, medical or financial.
Tips for a better transcript
- Choose the spoken language instead of “Detect automatically” when you know it. Detection listens to the first speech and can pick a related language.
- For Indian languages, use Whisper Small, or Large v3 Turbo on a computer with WebGPU.
- Clean up a noisy recording first with the noise reduction tool, then transcribe the cleaned file.
- Pick “English translation” to get English text from Hindi or another language in one step (Tiny, Base and Small can translate).
Subtitles: SRT and VTT
The SRT file works with YouTube, VLC, most video players and editing apps; VTT is the format web video players use. Long lines are split into caption-sized pieces. Timestamps come from the model and can be half a second early or late, so check them in a player. To burn captions into a video, or to fix them on a timeline, use Auto captions in the video editor, which uses the same models.
Frequently asked questions
Is my recording uploaded anywhere?
No. The file is decoded and transcribed inside your browser. The only download is the speech model, from Hugging Face, the first time you use it; after that it is kept in your browser and works offline.
Which speech model should I choose?
Whisper Tiny is quick and fine for clear English. For Hindi and other Indian languages choose Whisper Small, or Large v3 Turbo if your device has WebGPU (recent Chrome or Edge on a computer). The bigger the model, the better the accuracy and the longer the download.
Can it transcribe Odia?
No. Whisper has no Odia. It covers Hindi, Bengali, Telugu, Marathi, Tamil, Urdu, Gujarati, Kannada, Malayalam, Punjabi, Assamese, Nepali and Sanskrit among its languages.
How long can the recording be?
There is no fixed limit; an hour-long lecture works on a computer. It is held in memory while it is transcribed, so on a phone, recordings up to about half an hour are the safe range.
Which files can I use?
Any audio or video file your browser can play: MP3, M4A, WAV, OGG, Opus, FLAC, MP4, MOV, WebM and more, including WhatsApp voice notes. For a video, only its sound is used.