On-device speech recognition
WhatsApp voice note to text
Read a voice message instead of listening to it — in a meeting, on a train, or when it is eleven minutes long and you only need one sentence. The speech model runs on your own device, so the recording is never uploaded.
Drop an audio or video file
MP3, WAV, M4A, OGG, MP4, WebM — anything your browser can play
Why people want voice notes as text
Voice messages are fast to send and slow to consume. A four-minute note holds maybe thirty seconds of information, and you cannot skim it, search it, quote it in an email or paste it into a document. Family groups, contractors, doctors and small-business chats now run almost entirely on them, which means the important details of someone's week live in audio nobody can index.
Transcription solves that, but the usual services want an upload — and voice notes are personal by nature. A medical instruction, a payment agreement, a family argument: none of those belong on a stranger's server with an unread retention policy.
How the local model works
A compact Whisper speech model is downloaded to your browser cache once, then executed by your own hardware — WebGPU where the device supports it, otherwise WebAssembly across several CPU threads. The audio is decoded, resampled and fed to the model entirely inside the tab. After the first download it works fully offline, and the tool picks precision and thread count to match your device rather than assuming a desktop.
Need the recording itself in a normal format? The WhatsApp audio to MP3 converter handles that side, with the same no-upload guarantee.
Step by step
1. Save the voice message
Long-press the message, choose Share, then Save to Files on iPhone or Downloads on Android. On WhatsApp Web, hover the bubble and click the download arrow.
2. Drop it into the transcriber
OPUS, OGG, M4A, MP3, WAV and video files are all accepted. The recording never leaves your device.
3. Choose the spoken language
Picking the right language matters more than anything else for accuracy — a Hindi note transcribed as English produces nonsense. Leave it on auto only when you are unsure.
4. Read, copy or download
The text appears as it is produced, and can be copied out or saved as a plain .txt or .srt file with timings.
Questions people ask
- How accurate is it on a WhatsApp voice note?
- Voice notes are the hardest case for any speech model: a phone microphone at arm's length, a low bitrate, background noise, and often two languages in the same sentence. With the right language selected, clear speech transcribes very well; a noisy market recording will need cleanup. Read the result before you forward it as a record of what was said.
- Does it work for Hindi and other Indian languages?
- Yes — Hindi is one of the supported languages, alongside English, Spanish, French, Portuguese, Arabic, Bengali, Russian, German, Indonesian, Japanese and Chinese. Choose the language explicitly rather than relying on detection; Hinglish notes usually do best on Hindi.
- Is the audio sent to a server?
- The speech model itself is downloaded to your browser the first time and runs on your device's own processor or GPU, so the recording stays local. On devices too limited to run the model, the tool offers an optional assisted mode and asks first — you are never opted in silently.
- Why is the first run slow?
- The model files are a few hundred megabytes and are cached after the first download, so run one is the only slow one. After that, transcription starts instantly and works with no connection at all.
- Can I get subtitles with timestamps?
- Yes. The transcriber can export .srt with per-line timings, which is useful when you need to quote a specific moment in a long note or attach captions to a shared clip.
- I just need the audio in a playable format.
- Use the WhatsApp audio to MP3 converter instead — same privacy model, but it re-encodes the file rather than reading the speech.