All narration is generated locally on the user's machine without sending text or audio to external servers.
Yes. Individual narration segments can be regenerated independently.
Yes. Narration timing is derived directly from subtitle or timeline data.