Utilities
Transcribe audio to text
Choose an audio file or record from the microphone, and get text you can edit, copy or download. The speech model downloads once and stays on your device.
This tool didn't load
Nothing was sent anywhere — whatever you added stayed on your device. Open the full tool, or reload the page.
Save and continue on the desktop
Download or copy your result here first. Open the free desktop to save to Files or use more features.
- Subtitles (.srt, .vtt) and timestamps in the full Transcriber
- Save to Files when you continue on the desktop
- Every tool in one place, no new tabs
After you download, continue on the desktop for Files or the full app.
What Audio Transcriber can do
It transcribes on this page
Choose an MP3, WAV, M4A, OGG, WebM or FLAC file, or record from the microphone, and pick the language spoken. The list starts at your browser's language, because the model does not detect it. The speech model, Whisper tiny, downloads once and stays cached: 40 MB on a phone or without WebGPU, 150 MB on a computer with WebGPU. The transcript is editable: fix a misheard name, and Copy and the .txt download take your edits.
Limits, and why a phone gets less
Up to 100 MB and 30 minutes on a computer, 50 MB and 10 minutes on a phone: the model runs in the browser's memory, and a phone has less of it. An hour-long lecture has to be cut into parts first. Desktop Chrome or Edge is the fastest; a phone can be slow. Treat the result as a draft: names, abbreviations and terms are what a small model mishears most, especially on a noisy recording. It is built for speech, not songs, and a video has to give up its audio track first.
What the full app adds
The Local Transcriber in nopa adds a larger model for noisy recordings and technical vocabulary (Whisper base, 75 MB, or 290 MB with WebGPU), timestamps you can click to jump through the audio, telling speakers apart, and subtitles as .srt or .vtt. It runs on the same device; the recording is not sent anywhere.
Related apps
Questions
Is my audio uploaded?
No. The audio is decoded and transcribed in this browser tab. There is no server version to fall back to: if the browser cannot run the model, the transcription does not start.
What languages are supported?
Fourteen: English, Spanish, German, French, Portuguese, Italian, Russian, Ukrainian, Polish, Turkish, Indonesian, Chinese, Japanese and Korean. Pick the one spoken in the recording before you start; the page does not detect it.
Does it add timestamps or split speakers?
Not on this page, which gives plain text. The desktop app adds both, with .srt and .vtt subtitles; telling speakers apart downloads a 26 MB voice model once.