First use downloads a voice model of roughly 100 MB. Keep this tab open while it works. Long documents can take a while; a desktop computer works best.
Privacy and document support
Your document text and OCR stay in this browser. Voice-model, OCR and language files download from their open-source CDNs. Your text is not sent to a narration service.
For scanned English PDF pages, use the OCR button and review the recovered text carefully. Columns, handwriting, unusual typefaces, footnotes, tables and equations may need correction. Images are not described automatically. For a web article, paste its text above.
Narration: Kokoro (Apache 2.0). OCR: Tesseract.js (Apache 2.0). MP3 encoding: LAME via wasm-media-encoders. Open-source notices.