Whisper transcription online — free, unlimited, in your browser
OpenAI's Whisper, running client-side with WebGPU acceleration. No install, no Python, no API key — and unlike most Whisper sites, no file-length cap. Drop a 2-hour WAV and it just runs.
Free plan today: WAV files of any length; MP3/M4A up to 1 hour per file for now.
Most "Whisper online" sites cap you. This one doesn't.
Running Whisper in the browser has the same cost structure for every site that does it: the user's device does the work, so there's no per-minute server bill. Despite that, most browser-Whisper tools still gate the free tier at a few minutes per file. We don't — the free plan's only real limit is the compressed-format cap below, and that's a decoder constraint, not a business model.
| Tool | Free limit per file | Runs locally |
|---|---|---|
| Record to Notes | WAV: unlimited (2h+ tested) · MP3/M4A: 1h | Yes — WebGPU, WASM fallback |
| 1Transcribe | Files under ~5 minutes | Yes |
| Whisper Web (whisperweb.net) | ~20 minutes / 200MB | Yes |
| Typical hosted Whisper APIs | Metered per minute, key required | No — server-side |
Competitor limits as of September 2026 — they change, check each site. What doesn't change: if a tool runs Whisper on your device, a tight free cap is a choice, not a cost.
What's actually running
- Models: the open-source Whisper models — Tiny (~40MB, fastest), Base (~80MB, balanced), Small (~250MB, most accurate) — converted to ONNX and quantized to run in-browser. Downloaded once, then cached.
- Runtime: WebGPU where your browser supports it; a WebAssembly fallback everywhere else. The fallback is slower but produces the same output.
- Speed: about 10 minutes for a 2-hour file on a recent laptop with GPU acceleration. Output streams live, so you're reading the transcript while it runs.
- Privacy: transcription makes zero network calls after the model downloads — watch the network tab, or go offline mid-file. The audio never leaves your machine.
- On top of raw Whisper: auto-detected chapters with per-chapter summaries and timestamps, exported as TXT, SRT or Markdown — the part plain Whisper doesn't give you.
When you'd want this instead of running Whisper yourself
If you're comfortable with pip install and have a GPU, local Whisper is great — this is the same thing minus the setup: no Python environment, no CUDA drivers, no model-download scripts, no command line. Same models, running in a tab, with chapters and export built in. And if you're on a work machine where you can't install anything, a browser tab is the whole point.
The trade-off is honest: a browser can't use your GPU as fully as native CUDA can, so a local install with a strong GPU will beat it on speed. For everyone who doesn't want to maintain that setup, this is the shortest path to Whisper-quality transcription. For long recordings, see lectures and audio to notes; for short phone recordings, see voice memo to text.
Pro — when local Whisper isn't enough
- Cloud speed: a 2-hour file in about a minute (vs ~10 minutes locally)
- Highest accuracy in 90+ languages — beyond what the Small model does in noisy audio
- Batch queue: a whole folder of recordings in one go
- Study Packs: structured notes, glossary, quiz, one-click Anki export
FAQ
Is this really Whisper, or a lookalike?
It's the open-source Whisper models (Tiny, Base, Small), converted to ONNX and quantized so they run in a browser. Same architecture, same weights family — packaged to run client-side instead of on a server.
Why is it free with no cap when other Whisper sites charge?
Because your device does the compute, our cost doesn't grow with your minutes. Sites that cap browser-based Whisper are choosing to, not covering a bill. Our paid tier exists for cloud speed and accuracy, not to meter the local mode.
Do I need a GPU?
No. With WebGPU support (recent Chrome and Edge) transcription is GPU-accelerated; without it, a WebAssembly fallback runs on the CPU — slower, same results.
Is my audio uploaded to a server?
No. After the model file downloads, transcription runs with zero network calls — verifiable in your browser's network tab. On the free plan your audio never leaves your computer.
How long can files be?
WAV: no limit — 2-hour files are routine. MP3/M4A: up to 1 hour per file for now, with longer compressed-file support planned.
Which Whisper model should I pick?
Small (~250MB) for best accuracy, Tiny (~40MB) when speed matters more, Base in between. They download once and cache. For clear English, Small is usually the sweet spot.
What do I get beyond the raw transcript?
Auto-detected chapters with summaries and timestamps, plus TXT, SRT and Markdown export. That's the layer plain Whisper doesn't do — the transcript arrives already structured.