Record from mic
Tap to start
No files yet. Drop audio or video files above, or record from mic.
Idle.
How this works: Whisper runs entirely in your browser via ONNX Runtime + WASM. First run downloads the model (~40–500 MB depending on size); every file after that reuses the cached model with no network activity. Your audio never leaves your device — DevTools → Network during a run shows zero POST requests. Model files fetch from HuggingFace CDN on first use only.
Bulk mode: Drop as many files as you want. Each row shows its own progress and, once done, its own Copy and Download buttons. "Copy all" concatenates every transcript with file-name headers. "Download ZIP" packs every transcript into a single archive.
Practical notes: tiny/base = fast on any laptop small = ~2× real-time on M-series / good GPUs .en variants = 15% smaller, English only. Long-form audio (over ~10 min) is chunked automatically; total time is roughly 0.5–2× the audio duration on a modern machine.
Bulk mode: Drop as many files as you want. Each row shows its own progress and, once done, its own Copy and Download buttons. "Copy all" concatenates every transcript with file-name headers. "Download ZIP" packs every transcript into a single archive.
Practical notes: tiny/base = fast on any laptop small = ~2× real-time on M-series / good GPUs .en variants = 15% smaller, English only. Long-form audio (over ~10 min) is chunked automatically; total time is roughly 0.5–2× the audio duration on a modern machine.