Philosophy

English · 简体中文

I build Zotero-OpenReader for the way I read, and step by step it is moving off Zotero's Read Aloud.

Why it exists

I use Zotero-OpenReader every day to listen to papers and novels. Zotero 10's Read Aloud makes design choices and trade-offs that do not suit me. The plugin began by adding voices to it and now changes it wherever I want it to work differently; step by step, it will stop depending on Read Aloud.

The yardstick

• My preference decides. A feature is built because I want it.
• Requests are welcome. I build the ones I like that do not get in my way; when I will not build one, I say so and close it.
• Bugs are fixed by severity, whoever hits them. On a setup I cannot reproduce, I ask the reporter to confirm the fix.

Rules

1. Honest signals. A voice that does not report word timings highlights by sentence; the plugin never invents timings. Network calls time out and report; they do not hang.
2. Bring your own provider. No vendor is required or hard-coded; "OpenAI" means any server that speaks that API. Keys go only to the provider they belong to and to the backups you choose.
3. No silent spending. The plugin never spends your money without you knowing.
4. Uninstalling is safe. Remove the plugin and Zotero keeps working.
5. Every hook written down. Where the plugin reaches into Zotero's internals, each hook is verified against Zotero's source, recorded in notes/NOTES.md and pinned to one Zotero major.
6. A setting must do something. A setting that turns out to have no effect is removed, as the synthesis speed was.

What stays out

• My own analysis of the document. Telling body text from headers, footers and citations, and following a paragraph across columns and pages, stays Zotero's: Zotero maintains it for more than Read Aloud, and redoing it would be a project of its own.
• Features that fail silently when Zotero changes.