Text to Speech — read any text aloud in your browser

The tool is a thin React layer over the SpeechSynthesis interface that every modern browser exposes. There is no audio model shipped with this page — when you click Speak, a SpeechSynthesisUtterance is constructed and handed to window.speechSynthesis.speak(...), and your operating system's voice engine renders the audio. Privacy follows from the architecture: the browser never sends the text to a remote server unless you have picked a voice that reports localService = false, and the voice picker labels every such voice as "cloud" for transparency.

Three transformations sit between the textarea and the engine:

1. Sanitisation. The input is normalised — bidi override marks and zero-width joiners are stripped (they make some engines skip the surrounding word), newlines are replaced with". "so the engine pauses at line breaks, and any double-space run is collapsed.
2. Parameter clamping. The W3C spec defines the valid ranges for rate (0.1–10), pitch (0–2), and volume (0–1). Anything outside the range is clamped before it reaches the engine; anything non-numeric falls back to the default of 1. The "Verified · clamped to spec" badge in the tool header is computed live from a self-check that the clamp functions satisfy clampX(clampX(v)) === clampX(v) for every sample input.
3. Chunking. Chrome ships a watchdog (Chromium bug 679437) that halts a single utterance after roughly fifteen seconds. The tool side-steps the watchdog by splitting the input on sentence terminators (.!?) into sub-utterances of at most 200 characters, queueing them via onend so the audio sounds continuous to the listener.

Voices are enumerated via window.speechSynthesis.getVoices(). The list is populated asynchronously by Chromium-based browsers, so the picker subscribes to the voiceschanged event and refreshes when the engine reports new entries. Voices are grouped by BCP-47 language tag (parsed per RFC 5646), with English (UK/US/AU/IN), Sinhala (LK), and Tamil (LK/IN) lifted to the top of the list to suit the induwara.lk audience.

A pre-flight duration estimate is shown below the textarea. It comes from the heuristic characters ÷ 5.1 ÷ 165 wpm ÷ rate— 5.1 characters per English word (Mayzner 1965, re-validated by Peter Norvig's Google Books corpus analysis) and 165 words per minute (midpoint of the MDN-documented platform default range of 150–180 wpm). The estimate lands within roughly ±10% on Chrome and Safari at rate=1.

• W3C — Web Speech API (Synthesis section)
• MDN Web Docs — SpeechSynthesis interface
• MDN Web Docs — SpeechSynthesisUtterance (rate / pitch / volume)
• Can I use — speech-synthesis API support matrix
• IETF — BCP 47: Tags for Identifying Languages
• Chromium issue 679437 — speak() halts after ~15 seconds
• Peter Norvig — letter frequency & mean English word length

The clamp ranges, chunking thresholds, and language-tag handling on this page were last cross-checked against the upstream specs on 2026-05-11. The page is reviewed whenever a Chromium voice-list regression lands or the WICG Speech API spec ships a new draft. If you spot an engine behaviour that disagrees with the methodology above, please email me below.