Text to speech: read text aloud and download MP3

Natural voices without an account

Paste up to 5,000 characters, choose one of ten voices and press Generate. You can listen right away and download the audio as MP3 or WAV for videos, presentations, podcasts, e-learning or to listen to an article on the go.

• English (American): Heart, Bella, Michael and Fenrir.
• Brazilian Portuguese: Dora, Alex and Santa.
• Spanish (Latin American): Dora, Alex and Santa.
• Speed: 0.8× for language learning, up to 1.5× to save time.

The voices come from Kokoro, an open model with 82 million parameters that sounds far more natural than the robotic voices built into most systems. The first time, the page downloads the model (about 100 MB); after that it is cached and speech starts in seconds.

Tips

• Punctuation shapes the reading: commas add short pauses, full stops longer ones.
• Write numbers as digits: "1,250" and "$3.50" are read in words.
• Use the voice that matches the language of the text.

The audio is yours to use; the model's licence (Apache 2.0) allows commercial use.

Each sentence is converted to phonemes, the sounds of the language, using a pronunciation dictionary for English and spelling rules for Portuguese and Spanish (Latin American pronunciation: "c" before e/i and "z" sound like "s"). The Kokoro model turns those phonemes into audio in the chosen voice, sentence by sentence, on your device.