Are Text-to-Speech Apps Safe? A Privacy Checklist

Cloud speech generation sends the text being synthesised to a service. Depending on the product, that might be a sentence, a chapter, or an uploaded document. Do not assume every service keeps a permanent copy of the entire file: its processing and retention policy should explain that. Some apps offer both local and cloud voices, so check the voice you actually selected.

With on-device synthesis, the speech model can turn text into audio locally. Other features may still upload documents for syncing, conversion or backup. Ask about those features as well as the voice. The same distinction applies to a scanned PDF: finding text in an image and reading that text aloud are separate operations.

1. Processing: which features send text, document images or audio to a service?
2. Storage: is content kept only for processing, saved to an account, or included in backups?
3. Other uses: does the provider describe analytics, model improvement and third-party recipients separately?
4. Controls: can you remove uploaded content, disable optional collection and close an account?
5. Device access: do requested permissions fit the feature, such as camera access for scanning?

Apple's App Store privacy information is supplied by developers. It is a useful starting point alongside the full policy, rather than independent proof of every behaviour. A short policy is not inherently better than a detailed one. If a point matters to your use and is unclear, ask the provider before supplying the document.

No. It can show that a particular passage plays without a current connection. Audio may have been cached earlier, and an app may queue diagnostic events to send later. Failure offline can also mean a missing voice download or a licence check; it does not prove your document was uploaded.

For a useful availability check, install the app and desired voice, disconnect Wi-Fi and cellular, then import a new, non-sensitive local file and try several sections. Treat the result as evidence about that workflow. On iPhone and iPad, Apple's App Privacy Report can show contacted domains, but a domain list alone does not tell you the contents of each request.

LoudReader imports books into local app storage and generates narration on the device. It is an iPhone and iPad app that also runs on compatible Apple Silicon Macs as an iPad app. No LoudReader account is required. This removes a speech-server upload from the reading workflow.

It does not remove every network connection. Version 1.12 sends TelemetryDeck usage analytics, enabled by default, and Sentry crash and performance diagnostics. Its Settings screen does not currently expose an analytics switch. The app is designed to exclude reading text from those reports and scrubs diagnostic paths and URLs; that is not a promise that software can never contain a privacy defect. Downloads and App Store purchases involve other services too.

Read the privacy policy for the disclosure, and use the confidential-document checklist when device permissions, backups and where you listen matter as much as speech generation.