Utilore

Text to Speech

Paste the text, pick a voice and press the button. Everything is synthesised inside the browser: the system voice starts instantly, the neural voice needs a one-time download but sounds the same on every device and can be saved as a file.

  • 0Characters
  • 0Approximate length
  • 0Voices available

The system voice cannot be saved: the browser sends it straight to the sound card without giving the page an audio stream. The neural voice is built on your device, so it can be downloaded as a WAV file.

How to use

  1. Paste or type the text to be read.
  2. For an even voice on any device — and a file you can keep — switch on the neural voice: the model downloads once and stays in the browser.
  3. Choose a voice — the list comes from the languages installed on your device.
  4. Set speed and pitch if the default reading is too fast or too flat.
  5. Press «Read aloud»; you can pause and continue from the same place.

Good to know

Where the voices come from

The browser does not carry voices of its own — it asks the operating system. That is why the list differs from machine to machine: a phone usually has several voices per language, a desktop Linux may have none at all until a speech package is installed. It also means there are no character limits and no subscription: the work is done by software you already own.

Three voices, three trade-offs

The system voice costs nothing to start: it is already installed and speaks the moment you press the button — but how it sounds depends on the machine, and on a bare Linux desktop there may be no voice at all. The neural voices are downloaded models that run on your own device: Kokoro, at eighty-eight megabytes, sounds the liveliest and covers this language; Piper is lighter at sixty and knows every language on the site. Either way the text stays with you and the result can be saved as WAV.

Long text is read in pieces

Handed a whole article, the speech engine gives up part way: Chrome stops after roughly fifteen seconds unless the queue is nudged, and Safari drops everything past the first few hundred characters. So the text is cut at sentence boundaries and queued one phrase at a time. Cutting by character count instead would put the break inside a word, and that is audible as a stutter.

Frequently asked questions

Can I download the audio as a file?

Yes, with the neural voice: it builds the audio on the page itself, so a «Download WAV» button appears next to the player. The system voice cannot be saved — the browser sends it straight to the sound card without giving the page an audio stream, so there is nothing to assemble a file from.

There are no voices in the list. What now?

The system has none installed. On Windows they are added under Language settings, on Android and iOS under Accessibility, on Linux by installing a speech-dispatcher voice package. Reload the page afterwards — the list is read at load time.

Why does the same text sound different in another browser?

Because the voice comes from the system, but the choice of default voice, the handling of pauses and the exact pace belong to the browser. Chrome may also offer network voices, which sound smoother and need a connection.

Is the text sent to a server?

The page sends no text anywhere. The neural voice downloads its model from huggingface once and then works offline; the words you paste never leave the device. A system voice marked as a network one may ask the vendor for audio, which is why local voices are preferred here.

Related tools