The app works from structured line data rather than from a raw PDF. Each line needs a speaker and spoken text. That allows the browser to know which lines belong to your selected role, which lines are cues, and how to jump accurately through the scene.
The selected character becomes your part. All other lines remain visible as cue lines and can be played back with browser speech synthesis. This is what turns a shared script into a one-person rehearsal tool.
If the current line belongs to another character, the app can play it and advance. If the current line belongs to you, the app can wait for your delivery, either through manual speaking or automatic listening where the browser allows it.
If a script import is messy, you fix it in the cleanup editor. If a rehearsal run is good, you keep the cleaned file for later. The tool is meant to get better over repeated use of the same cleaned rehearsal material.
The browser speaks cue lines using available local voices. Different devices sound different because the voices come from the browser and operating system.
When supported, the browser listens to your line and compares it with the expected text. This is helpful for rehearsal feedback, but it is still approximate.
The app itself runs as a static site. There is no account login, no scene project system, and no server-side script library attached to your sessions.