Quick Answer
To transcribe audio privately for free, open this tool in Chrome or Edge, drop in an MP3, WAV, M4A, or most video files, and press Transcribe — an open-source OpenAI Whisper model downloads once (~40–250 MB), then runs entirely inside your browser, producing a timestamped transcript you can copy or export as TXT, SRT, or VTT. Your recordings are never uploaded to any server, which makes it a free alternative to MacWhisper Pro and paid transcription services for confidential material.
How It Works
Pick an audio or video file — MP3, WAV, M4A, OGG, FLAC, or most MP4/WebM/MOV videos (Chrome extracts the audio track). Drag it in or click to browse; the file stays on your device
Choose a Whisper model — tiny.en (~40 MB, fastest, English only), base (~80 MB, balanced, multilingual), or small (~250 MB, best accuracy) — and a language if you do not want auto-detection
Press Transcribe: the model downloads once from the CDN and then runs entirely in this browser tab (WebGPU-accelerated where available), producing a timestamped segment table
Review and export: edit the full transcript inline, click any segment to seek the audio player, then copy the text or download TXT, SRT, or VTT subtitle files
Key Facts
- 100% on-device: OpenAI Whisper runs in your browser via WebAssembly (or WebGPU when available) — audio never leaves your machine
- Accepts MP3, WAV, M4A, OGG, FLAC, and most MP4, WebM, and MOV videos (Chrome extracts the audio track)
- Timestamped segment table with click-to-seek playback synced to an inline audio player
- Editable full-transcript box with one-click copy to clipboard
- Download as plain TXT, SRT subtitles, or WebVTT files with standard HH:MM:SS,mmm timecodes
- Three model sizes: tiny.en (~40 MB, English), base (~80 MB, multilingual), small (~250 MB, best accuracy)
- Auto-detect or choose from 17 languages on the multilingual models
- Free, unlimited, no sign-up, no watermarks — works offline after the first model download
Frequently Asked Questions
Is my audio really never uploaded?
Correct. The AI model (OpenAI Whisper, converted to run in browsers) is executed by WebAssembly directly in your browser tab. The only network traffic is the one-time download of the open-source model files themselves — your recordings are decoded locally by the Web Audio API and never transmitted, stored, or seen by any server. After the first load the model is cached and the tool keeps working even if you go offline.
What audio and video files can I transcribe?
Any audio format your browser can decode: MP3, WAV, M4A/AAC, OGG, Opus, and FLAC. For video, Chrome and Edge extract the audio track from most MP4, WebM, and MOV files, so you can drop recordings straight in. Safari and Firefox are pickier about video containers — if decoding fails, export the audio as MP3 or WAV first. There is no file-size cap; longer files simply take more time on your own hardware.
How large are the AI models, and do I download them every time?
tiny.en is about 40 MB, base about 80 MB, and small about 250 MB. You download a model only the first time you use it — the browser caches the files in its Cache API storage, so every later run starts instantly and works offline. Switching models downloads that model once as well; each cached model stays available for future sessions.
How accurate is browser-based Whisper transcription?
The same Whisper architecture that powers paid transcription services runs here, so accuracy is comparable for clear speech. The small model is closest to cloud accuracy, base is a good speed/quality balance, and tiny.en is best for quick English drafts. Noisy recordings, heavy accents, and overlapping speakers reduce accuracy on any Whisper implementation — when that happens, step up a model size. Segment timestamps are accurate to roughly a second, which suits subtitles and note-taking.
Is this transcription tool really free?
Yes — completely free with no sign-up, no credits, no watermarks, and no usage limits. Because the model runs on your own hardware instead of a server, there are no per-minute costs to pass on. Desktop apps like MacWhisper Pro and most transcription SaaS plans charge for exactly this kind of local or metered processing; this tool undercuts them at free by running Whisper in your browser.