How does browser speech synthesis work with on-device voices?
Desktop browsers often bundle remote neural voices that transmit every word to a third-party speech server over the internet. Filtering the browser voice list so only on-device voices are shown keeps confidential contracts, unreleased drafts, and personal notes strictly inside your tab with zero network traffic.
Filtering remote cloud voices with localService
Under the MDN SpeechSynthesisVoice localService property, browsers expose a boolean flag showing whether a voice runs locally on your operating system or on a remote cloud server. When the page loads, it calls speechSynthesis.getVoices() and keeps only entries where localService is true.
Any remaining remote-only voices are counted and excluded from the Voice dropdown. If a desktop browser has zero local voices installed, the player disables Play and explains that online voices were skipped to protect your privacy.
Android Chrome is the single exception handled in code. Because Android reports system-installed TTS voices with localService set to false even when synthesis runs on the phone itself, the tool falls back to listing Android system voices instead of showing an empty menu.
Installing extra system voices on Windows, macOS, and mobile
Operating systems ship with one or two basic voices by default, while higher-quality offline voices are optional downloads. On Windows, open Settings, Time & language, Speech, and install additional voice packs for your preferred locale.
On macOS, open System Settings, Accessibility, Spoken Content, and click Manage Voices to download Siri or Enhanced local voices. If you work with foreign-language documents, you can also translate foreign-language PDFs before listening in a separate tab before switching back to an English voice.
- Voice sorting order: Available voices are sorted automatically so exact matches for your browser
navigator.languageappear first, followed by same-language regional variants. - Fallback timer: Because some browsers load voices asynchronously, a 700 ms timer re-checks
speechSynthesis.getVoices()if thevoiceschangedevent fires late.
Sentence splitting, speed control, and Chromium utterance limits
Pasting a long report into a naive speech engine often breaks mid-abbreviation or stalls completely after a few sentences. This reader normalizes whitespace, protects common abbreviations, and chunks long unpunctuated clauses before handing them to the browser speech queue to read aloud.
Preserving abbreviations, numbers, and bulleted lists
When you click Read this, the parser splits paragraphs on blank lines and detects whether single-newline blocks represent bulleted or numbered lists. Short lines or items starting with list markers stay separate so each bullet reads as its own unit instead of running together.
Sentence detection keeps titles and acronyms such as Dr., Mrs., e.g., i.e., and U.S. intact when followed by lowercase words or mid-sentence continuations. Each resulting sentence becomes an interactive span in the reading pane.
Before listening to a long essay or script, you can check draft word count and reading time before playback to compare silent reading pace against audio duration. The player itself estimates listening minutes using a baseline of 160 words per minute divided by your chosen speed multiplier.
Preventing 15-second stalls with 200-character sub-chunks
According to the MDN SpeechSynthesisUtterance interface, each utterance queues a single text string with a selected voice, language tag, and playback rate. Desktop Chromium browsers have a well-known engine limit that silently pauses or drops any single utterance lasting longer than roughly 15 seconds.
To prevent playback from freezing on long legal or academic sentences, the player splits any sentence exceeding 200 characters into smaller sub-utterances. It searches backward from 200 bytes or characters for semicolons, colons, commas, or hyphens so pauses sound natural.
| Control or parameter | Verified code behavior | Practical effect during listening |
|---|---|---|
| Speed selector | Eight steps: 0.75×, 0.9×, 1×, 1.15×, 1.3×, 1.5×, 1.75×, 2× |
Updates live from the current sentence without restarting |
| Utterance chunk cap | subChunkUtterance() splits clauses over 200 characters |
Prevents desktop Chromium 15-second playback freezes |
| Duration estimate | wordCount / (160 * rate) rounded to nearest minute |
Displays approximate listening time beside the title |
| Voice filter | Requires localService === true (except Android Chrome) |
Blocks cloud voice APIs from receiving pasted text |
| PDF text floor | Rejects PDFs yielding fewer than 20 characters | Surfaces an actionable warning on blank or unreadable scans |
Navigating interactive sentence highlighting and PDF mode
Long proofreading sessions require jumping directly to problem paragraphs rather than listening from the top after every edit. The reading pane provides interactive sentence highlighting that tracks the active sentence, scrolls it into view automatically, and accepts both mouse clicks and full keyboard navigation.
Clicking sentences and using keyboard shortcuts to jump
Under the MDN Web Speech API SpeechSynthesis reference, changing voice or speed mid-stream requires canceling and re-queuing the active utterance. Whenever you change Voice or Speed during playback, the tool cancels the old utterance and resumes immediately from the current sentence index.
You can click any sentence span directly to begin speaking from that exact spot, or use Previous sentence and Next sentence to step backward and forward. Keyboard users can arrow through spans with ArrowLeft, ArrowRight, ArrowUp, ArrowDown, Home, and End, then press Enter or Space to start reading.
Switching from pasted text to scanned or digital PDFs
At the top of the workspace, the What to read segmented toggle lets you switch between Paste text and A PDF without leaving the page. Dropping a PDF extracts text page by page and runs client-side OCR automatically on scanned pages, displaying a notice listing which pages used OCR.
Multi-page documents unlock the Go to page dropdown in the player sidebar so you can jump straight to a specific chapter or appendix. If a PDF is too dense to listen to in full, run it through a browser-based PDF summarizer first and paste the condensed notes back here.
Tip: Click Download the text in the player sidebar after loading a PDF or editing pasted paragraphs to save a clean
.txtcopy with double-spaced paragraphs directly to your device.
Listening playback vs downloadable audio files
Many visitors look for an MP3 or WAV download button after pasting a script, only to find live playback controls instead. Understanding how browser speech synthesis differs from audio recording clarifies what the tool exports and how to prep text beforehand.
Why Web Speech API plays live without MP3 export
The W3C Web Speech API Specification routes synthesized speech straight to the operating system audio output device and intentionally withholds raw PCM samples from JavaScript. Because the browser provides no audio stream or buffer to capture, no purely client-side Web Speech tool can export MP3 or WAV files without sending your text to a remote server.
This design trade-off guarantees zero data leakage. Your text stays in local memory while your device speaker plays the voice in real time, and clicking Change clears the active queue immediately.
Cleaning messy drafts and exporting plain text
Text copied from PDFs, emails, or web pages often carries irregular capitalization, all-caps headings, or stray line breaks that make speech synthesis sound robotic. Running rough drafts through the Case converter to normalize sentence case or title case before pasting improves inflection and pacing.
Once you load a document or paste formatted text, clicking Download the text exports a clean .txt file named after your uploaded PDF or text.txt for pasted content. Every line of extraction, splitting, and playback happens inside your browser tab, so nothing is stored once you close the window.