PDF Read Aloud
What data does an AI PDF reader send to a speech provider?
It depends on the application. In PDF Read Aloud’s browser PDF workflow, the original PDF bytes stay in your browser. The text requested for narration goes to the Insightable backend and then to OpenAI for AI speech generation. The next section can be sent in advance for preloading. Local PDF viewing therefore does not mean all processing is local.
Which parts are processed locally and which use a server?
The browser renders PDF pages, extracts text and maps your selection to narration sections. To generate audio, it sends section text and the selected voice, delivery style and language through the authenticated Insightable API. OpenAI returns speech audio. Replaying a cached section avoids another generation request; changing text or voice can create a new request.
Does selecting one sentence send only that sentence?
Not necessarily. By default, a selection sets the starting point and narration continues into later text. Continuous reading can prefetch the next section before the current audio finishes. Enable “Read only the marked text” if you want to limit the reading range, and review the editable section before generating. Do not infer data scope from the visible selection alone.
Is text or audio stored, and is this a zero-retention service?
This prototype caches a limited number of generated clips in browser and backend process memory. It also stores listening-position metadata and voice settings in browser local storage. This is not a zero-retention claim: the provider’s logging and retention terms apply separately. OpenAI documents that API data is not used for training by default unless you opt in, but that is distinct from retention and from eligibility for special data controls.
Can I use confidential research or personal data?
Do not assume this prototype is approved for confidential manuscripts, participant data or sensitive personal information. Check permissions, your institution’s policies and the current processing agreements and provider terms first. For a voice-quality trial, use a public PDF you may process, ideally a short, non-sensitive document.
More frequently asked questions
Does the app expose the OpenAI API key to my browser?
No. Speech generation is proxied by the backend, where the provider key remains. You use your Insightable app account rather than entering a provider key into the reader.
Is the generated voice human or AI?
The narration is AI-generated speech. Natural-sounding delivery does not make it a human recording, and pronunciation or emphasis can still be wrong.
Sources and scope
Product descriptions refer to the current prototype. General technical explanations are supported by the primary sources below. Provider terms can change.