voice·clone

Record your own voice-cloning dataset in the browser, with a teleprompter, a level meter, and a check on every take

Audio never leaves this device. The page writes into a folder you pick — there is no upload and no server.

1 · Working folder

Pick the folder holding script.md and raw/. Takes are written straight into raw/, numbered from the ones already there.

Voice Clone

Record your own voice-cloning dataset in the browser, with a teleprompter, a level meter, and a check on every take

Training a voice clone needs one to two hours of clean, consistent narration. The hard part isn't the recording — it's reading a long script without losing your place, and catching a bad take while you can still redo it rather than an hour later.

Point the tool at a working folder holding a `script.md` and a `raw/` directory. It renders the script as a teleprompter, records 48 kHz / 24-bit mono WAV with the browser's echo cancellation, noise suppression, and auto gain control switched off, and scores every take on peak and mean level the moment you stop. Takes are numbered and written straight back into `raw/`.

Audio never leaves the machine. There is no upload, no server, and no analytics — the page holds a handle to a folder you picked, and that is the whole storage layer.