Texto a voz

How browser text-to-speech works

This tool reads your text aloud using the Web Speech API, a feature built directly into modern browsers. When you click Play, the page hands your text to the speechSynthesis engine, which uses voices supplied by your operating system — the same voices that power your phone's screen reader or your computer's accessibility settings. Nothing is uploaded to a server; the audio is generated on your own device. That also means the list of available voices, and how natural they sound, depends on your OS and browser, not on us. A Mac, a Windows PC, and an Android phone will each show a different roster of voices.

Because the voices come from the platform, quality varies. Some systems include high-quality "neural" voices that sound close to human; others fall back to older, more robotic ones. Voices marked with a star are your system default for that language.

The three controls that shape the voice

These map directly to the rate, pitch, and volume properties of the speech request. They're applied at the moment you press Play, so adjust them first, then play — changing a slider mid-sentence won't alter audio already in progress.

A worked example

Paste a paragraph you're proofreading. Pick a clear voice, set Rate to about 0.9× so you can follow along, leave Pitch at 1.0, and Volume at 1.0. Press Play. The progress bar fills based on an estimate of how long the passage takes (roughly fifteen characters per second, divided by your chosen rate), and an elapsed-seconds counter ticks up. You can Pause to stop on a word and resume from the same spot, or Stop to cancel entirely and reset. Hearing your own writing read back catches dropped words, clumsy phrasing, and run-on sentences that the eye skips over.

Common use cases

• Proofreading. Errors that are invisible when reading silently become obvious when spoken — a missing "the," a repeated word, an awkward clause.
• Accessibility and reading support. People with dyslexia, low vision, or reading fatigue can listen instead of read.
• Language learning. Hear pronunciation and rhythm by selecting a voice in the target language.
• Multitasking. Listen to an article or your notes while cooking, commuting, or doing chores.
• Pronunciation checks. Type a tricky name or term and let the system voice say it.

Tips and limitations to know

• Voices load asynchronously. On first visit the voice list can briefly show "Loading voices…" before your system reports them. If the dropdown looks empty, wait a moment or reload.
• Language affects pronunciation. Choose a voice whose language matches your text. An English voice reading French will mangle accents, and vice versa.
• It's not a recorder. The Web Speech API plays audio live but does not, by itself, save an MP3. To capture the audio you'd need separate recording software; this tool is for listening, not exporting files.
• Length and tab focus. Some browsers throttle or stop synthesis when the tab loses focus, and very long passages can hit engine limits. For a long document, break it into sections.
• The progress bar is an estimate. It's based on character count and rate, not the engine's exact timing, so it's a guide rather than a precise clock — which is why it pauses just short of 100% until the real "finished" event fires.
• Pitch and rate at extremes. Pushing Rate to 2.0× or Pitch to the edges quickly sounds unnatural. Small adjustments around 1.0 give the most usable results.

Getting more natural-sounding results

The single biggest factor in how human a voice sounds is which voice you pick — a system's neural or "enhanced" voice will always beat its legacy one, so scroll the full list rather than accepting the first option. Beyond that, the engine reads punctuation as prosody cues: commas create short pauses, periods longer ones, and question marks lift the intonation at the end. Writing in clean, well-punctuated sentences makes synthesis sound dramatically better than feeding it a wall of unbroken text. Spelling out abbreviations (Doctor instead of Dr.) and numbers when you want them read a particular way also removes guesswork from the engine.

Privacy

Speech is synthesised entirely on your device by your browser and operating system. The text you type is not sent to BoxTool or any third party — you can confirm this by going offline, since the built-in voices keep working without a connection (cloud-backed voices are the exception and depend on your OS).