AI Speech-to-Text (Whisper)

How it works

1. Step · 01

   Drop an audio or video file (MP3, WAV, M4A, WebM, MP4…).

2. Step · 02

   First run downloads OpenAI's open-source Whisper-tiny model (~80 MB). Cached in your browser after that.

3. Step · 03

   The audio is decoded and transcribed 100% locally — nothing is uploaded to any server.

Frequently asked questions

Which Whisper version?

Xenova/whisper-tiny (multilingual). Around 80 MB, fast enough for real-time on most laptops. For higher accuracy, we may add whisper-base / small in a future update.

What file types work?

Anything the browser can decode: MP3, WAV, M4A, AAC, OGG, FLAC, WebM and MP4 video (audio track is extracted automatically).

How accurate is it?

Whisper-tiny is good for clean speech in the supported languages. For noisy audio, heavy accents or long recordings, larger Whisper models (not yet shipped here) do better.

Is my audio uploaded?

No. Audio is decoded and transcribed 100% locally in your browser after the one-time model download.

Are my files uploaded to a server?

No. Files are processed directly in your browser using JavaScript and WebAssembly. Nothing is uploaded to our servers, so your documents never leave your device.

Does it work on mobile?

Yes. The tools run in any modern browser on desktop, tablet, or mobile (Chrome, Edge, Safari, Firefox). Very large files may be slower on low-end phones.

• How to Use Claude AI Pro Features for Free (2026)What Claude Pro actually adds, the legitimate ways to get more AI capability at no cost in 2026, and free browser tools that cover common paid-AI tasks.7 min read