Text to Speech
Turn any written text into natural spoken audio — with dozens of voices and full control.
Voice Settings
0 voices available
Voices come from your device and operating system — the selection varies by browser and platform.
Text-to-speech has crossed the line from robotic novelty to genuinely usable technology, and modern browsers ship it built in — no account, no API key, no upload. This tool uses the Web Speech API, which routes your text through the voices installed on your device: on Windows those are Microsoft's natural voices, on macOS Apple's, on Android Google's. That architecture matters for privacy — with device voices, the text never leaves your machine — and explains the quirks: which voices you see depends on your operating system, and quality varies from clearly synthetic to nearly indistinguishable from recorded speech.
How the tool works
Type or paste text, pick a voice from the list your device provides, adjust rate and pitch if you want, and press speak. The browser's speech engine synthesizes audio in real time — you can pause, resume, or stop mid-utterance. The voice list differs by device and installed language packs: a Windows machine with Arabic language support installed will offer Arabic voices, and adding voices happens in your operating system's settings, not in the browser.
The controls
Rate 0.1 – 10 (1 = normal; 0.8–1.2 is the useful range)
Pitch 0 – 2 (1 = normal voice pitch)
Volume 0 – 1
Voice list = speechSynthesis.getVoices()
→ device voices (offline, private)
→ some browsers add cloud voices (online)
Proofreading speed: ~0.9 · Skimming: 1.5–2.0Rate is the control worth learning. Slowing to 0.9 for proofreading gives your ear time to catch errors your eye slides over; pushing to 1.5–2.0 works for consuming familiar material quickly, and regular listeners adapt to surprisingly high speeds. Extreme values in either direction break the prosody model — below about 0.5 or above 3, most voices stop sounding like language. Pitch changes are cosmetic and mostly useful for distinguishing multiple readers.
What to know about text-to-speech
- 1Listening is a genuinely different proofreading channel. Your eyes skip words and auto-correct typos because reading is predictive; your ears do not. Hearing your own writing read aloud surfaces missing words, doubled words, and clunky sentences that survived multiple visual passes. Writers who adopt this rarely go back.
- 2Voice quality is a device property, not a website property. The same text sounds different on Windows, macOS, and Android because each supplies its own synthesis engine. If the voices you see are poor, install better ones in your OS settings — Windows' natural voices and macOS's premium voices are free downloads that dramatically outclass the defaults.
- 3Punctuation is the prosody API. Speech engines pause at commas, drop pitch at periods, and rise at question marks — text with sloppy punctuation sounds wrong even with a perfect voice. Conversely, listening to your text is a sharp test of whether your punctuation actually matches your intended rhythm.
- 4Numbers, abbreviations, and symbols are guessed. "Dr." may become doctor or drive; "2/3" may be a fraction or a date; "IV" may be four or an intravenous drip. Engines guess from context and sometimes guess wrong — for critical listening material, spell out anything ambiguous.
- 5For long listening sessions, slightly slow and monotone beats expressive. Highly expressive voices are engaging for minutes and fatiguing over hours; audiobook narrators are trained toward evenness for the same reason. If you use TTS to consume articles, a plain voice at 1.1–1.3 is the sustainable setting.
Frequently asked questions
Is my text sent to a server?
With device voices — the default on most systems — no: synthesis happens entirely on your machine, and the text never leaves it. Some browsers also list cloud voices (often marked as online or premium), which do send text to the vendor's servers for synthesis. If privacy matters for your text, choose a device voice; if the tool works with the network disconnected, the voice is local.
Why do the available voices differ between my devices?
Because the browser exposes whatever speech engines the operating system provides. Windows, macOS, iOS, and Android each ship different voices, and installed language packs extend the list. This is also why voice names look platform-specific. To get more or better voices, add them in your OS's language and speech settings — the browser list updates automatically.
Can I download the speech as an audio file?
Not directly — the Web Speech API deliberately plays audio without exposing the stream, so there is no built-in save button. Workarounds exist: system audio recording captures anything played, and dedicated TTS services generate downloadable files. For personal listening the live playback covers most needs; for producing audio assets, a file-generating service is the right tool.
Why does the voice mispronounce some words?
Synthesis engines map text to sound using pronunciation dictionaries plus letter-to-sound rules, and both fail on names, technical jargon, loanwords, and ambiguous abbreviations. There is no user-accessible fix in the browser API, but deliberate misspelling works as a practical hack — writing a name phonetically produces the right sound even though the text is technically wrong.
Does text-to-speech work for languages other than English?
Yes, for any language with an installed voice — the quality hierarchy generally follows market size, with major languages getting near-human voices and smaller ones more mechanical treatment. The voice must match the text's language: an English voice reading Arabic text letter-names its way through unusably. If your language is missing from the list, install its language pack at the OS level and the voices appear.