LESS REPLAY. MORE CLARITY.
Turn a recording into
words you can work with.
01 / THE RECORDING
One file · 50 MB maximum · up to 3 minutes · formats your browser can decode. Video must contain an audio track.
The first transcription downloads approximately 42 MB of Whisper Tiny model files from Hugging Face, plus the local speech runtime. Your browser caches them when storage permits. The recording is processed in a worker and is never sent for transcription.
Model: not loaded
02 / THE WORDS
Add a recording or try the speech sample.
Your editable transcript and approximate segment times will appear here.
Edit text, start/end times in seconds and optional speaker labels. Speaker labels are manual: this tool does not identify speakers automatically. Check names, numbers and timing against the recording.
Choose a recording your browser can decode. Whisper Tiny runs locally on the CPU through WebAssembly, with English, Russian or automatic language detection. Keep this page open while the model works. A desktop browser is recommended; processing may take longer than the clip itself.
Download an editable text transcript, or use SRT and WebVTT files in a video editor. Segment times are estimates. Whisper can make mistakes or produce text that was not spoken, particularly with silence, music or unclear speech. This tool does not automatically distinguish speakers or translate the recording.
The pinned Whisper Tiny model is downloaded from Hugging Face on first use. Model files may remain in the browser cache until your browser evicts them. Audio samples, transcription and exports stay on your device. Unload model releases the running worker; it does not erase cached model files.
The sample is an excerpt of President John F. Kennedy’s inaugural address, bundled from the Hugging Face Transformers.js documentation dataset. Selecting it runs actual speech recognition; no transcript is prefilled.
