Turn typed text into speech with 177 neural voices across 53 languages. Pick a voice and a pace, play it here, or export the result as a WAV file. The engine runs inside your browser tab, so your text never leaves your device.
Each voice model downloads once and stays in your browser's storage, so repeat runs are instant and later sessions work without a connection.
Slow a passage down to catch every syllable, or speed it up to skim. The slider moves in .1 steps so you can settle on a pace you can follow.
Download the speech as a file with the same voice you previewed — something the operating-system voices behind the Web Speech API cannot do.
The audio you hear is the audio you export: download a WAV in the same voice, ready for an editor, a slide deck, or offline listening.
Pick the language and the voice quality you need instead of accepting whatever single voice the operating system happens to ship.
Synthesis happens in the browser tab. Nothing you paste is uploaded, logged, or sent to a speech API.
Slow the voice down for study or proofreading, speed it up to skim a long document — from .5x to 2x in fine steps.
It changes the pace of the speech between .5x and 2x. Slow it down for language practice or for proofreading a section you keep glossing over, speed it up to skim a long document. A faster pace also produces a shorter — and therefore smaller — file, because the audio itself is shorter.
177 voices across 53 languages. English has the widest choice with 38 voices, followed by German and Dutch with 10 each, Spanish with 9, French with 7, and Russian with 4. Smaller language families usually have a single voice, which is why it is worth checking the exact voice before you build a workflow around it.
Start with a medium voice, which most languages have: at 22.05 kHz it is the tier where listeners stop noticing the synthesis. Choose x_low or low when you want the smallest download or you are on an older machine — they render at 16 kHz and sound more synthetic. The high tier is the best Piper gets, but a high-quality model can be four times the size of a medium one.
Because it is uncompressed audio: mono 16-bit PCM at the voice's sample rate. At 22.05 kHz that is about 44 kB per second, or roughly 2.6 MB for every minute of speech. The upside is that no compression stage sits between the model's output and your file, so the audio can go straight into an editor or video timeline.