How does on-device PDF speech playback protect privacy?
All text extraction, OCR, and speech synthesis run inside your browser tab using your operating system’s built-in voices, sending 0 bytes of document text to external servers. Instead of streaming confidential contracts or medical PDFs to a cloud text-to-speech API, the player strictly filters for offline synthesizers.
Filtering out remote network voices with localService
When loading available voices, the player checks the MDN SpeechSynthesisVoice.localService Property on every browser voice, re-checking after a 700 ms fallback timer if voices load asynchronously. Desktop browsers often bundle natural-sounding network voices whose localService flag is false, meaning they transmit every spoken sentence to a remote server.
Because those remote voices violate zero-upload privacy, they are excluded automatically on desktop browsers. On Android Chrome, where system TTS voices sometimes report localService as false even when running locally, the player safely falls back to your installed Android system voices.
Using device voices without MP3 server rendering
Playing audio directly through OS speech engines explains why there is no MP3 download button: browsers do not let web pages capture system speech output into an audio file. If you received an email attachment saved as an .msg file first, open it with an Outlook MSG email viewer to export a clean PDF before listening.
Listening directly in the browser also means zero server-side audio queue or file size cap, so you can open multi-chapter reports immediately.
Cleaning headers, scanning with OCR, and sub-chunking sentences
Raw PDF extraction often mixes body prose with repeating chapter titles, copyright footers, and page numbers that ruin listening flow. Before populating the reading pane, the text pipeline cleans page furniture and adapts long sentences for reliable browser playback.
Stripping running headers, footers, and page numbers
The parser inspects the first and last paragraphs on populated pages with 3 or more blocks. Any edge line shorter than 90 characters that lacks terminal punctuation and repeats across pages, as well as standalone page numbers, is removed automatically.
| Processing step | Code rule or threshold | Purpose during playback |
|---|---|---|
| Minimum text guard | At least 20 characters | Flags blank or unreadable scans |
| Header/footer filter | Under 90 chars, repeating | Skips page numbers and running titles |
| Utterance chunking | Max 200 characters | Prevents 15-second Chromium freezes |
| Listening estimate | 160 words per minute base | Updates duration as speed changes |
When a scanned page yields no embedded text layer, browser OCR runs automatically and displays a notice listing the scanned page numbers. If your reading material starts as an ebook rather than a PDF, convert DRM-free EPUB ebooks into PDF first to listen with full page navigation.
Preventing browser speech freezes on long clauses
Desktop Chromium engines frequently stall when a single MDN SpeechSynthesisUtterance Interface exceeds roughly 15 seconds. To keep academic papers and legal clauses flowing, any sentence longer than 200 characters is split at semicolons, colons, commas, or spaces into chained sub-utterances.
What playback speeds, navigation, and export options exist?
You can adjust playback speed across eight presets from 0.75x to 2x, jump directly to any page via Go to page, or click any sentence in the transcript to begin reading from that exact point. As each sentence plays, the active span receives a yellow highlight and scrolls smoothly into view.
| Control or shortcut | Action in player | Live behavior |
|---|---|---|
| Speed dropdown | 0.75x, 0.9x, 1x, 1.15x, 1.3x, 1.5x, 1.75x, 2x |
Applies immediately from current sentence |
| Previous / Next | Step backward or forward 1 sentence | Resumes playback if currently playing |
| Arrow keys + Enter | Move focus between sentences | Starts reading from focused span |
| Download the text | Exports cleaned .txt transcript |
Excludes stripped headers and footers |
Estimated listening time updates dynamically based on a 160 words-per-minute baseline divided by your selected speed. Following the W3C Web Content Accessibility Guidelines (WCAG) 2.2 focus and keyboard interaction patterns, every sentence span supports Home, End, and arrow-key traversal.
Clicking Download the text exports a plain .txt file containing only the cleaned paragraphs, making it easy to archive meeting notes or study transcripts alongside audio review. If a 100-page report is too long to listen to in full, run Summarize PDF first to distill key takeaways before starting audio playback.
Managing scanned pages, language voices, and pasted text
Switch What to read between A PDF and Paste text whenever you want to proofread an email draft or listen to copied notes without creating a PDF. Pasted lists separated by single newlines stay broken into individual readable items instead of merging into one run-on sentence.
Installing extra system voices for offline listening
Because the MDN Web Speech API SpeechSynthesis Documentation relies on voices registered by your operating system, installing higher-quality offline voices in macOS Spoken Content, Windows Speech, or iOS Accessibility settings immediately expands the Voice dropdown. Reloading the tab after adding an OS voice package makes the new voice selectable right away.
Tip: Changing Voice or Speed while audio is playing restarts cleanly from the current sentence without losing your place in the document.
Switching between PDF files and pasted text drafts
If a document is written in another language and you do not have a matching offline voice installed, read foreign-language documents in your native tongue before loading the translated file into the audio player. Once finished, click Change to clear the active transcript and load a new file or text snippet.