A natural voice that reads to you, on your device.

Paste, choose, listen.

Type or paste your text, or open a text, Markdown or Word file. Under the box you see how many words it has and about how long it will take to say. Then answer one question, what it is for: Listen to a document reads a little quicker than talking pace; Video voice-over reads at talking pace, evens the loudness and adds subtitles; Audiobook leaves longer pauses between paragraphs and writes a chapter for each heading.

Choose one of nine voices, six with an American accent and three with a British one, and press the play button beside any of them to hear it say a sentence first. Press Make speech and it starts reading at once, while the rest is still being made; Stop halts it within a sentence.

• Open Text to speech
• Guide: text to speech with a British accent

MP3, WAV and subtitles.

When it is ready, the page says how long it came to, how many words and which voice, and Download saves it. The MP3 is written by the same writer as the site's audio tools, with LAME and a gapless header, and opened again and checked before it is offered; WAV is 16-bit and lossless. Subtitles come as SRT or WebVTT, each sentence at its exact time, at most two lines of 42 letters and seven seconds each.

For a video, the voice-over is measured as broadcasters measure loudness (ITU-R BS.1770) and set to −16 LUFS, the level Apple Podcasts asks of spoken word. For an audiobook, each heading of a Markdown or Word file becomes an ID3 chapter in the MP3.

• Guide: make audio an even loudness
• Guide: turn text into an audiobook

How the voice works.

Your text is cut into sentences, and numbers, money, times and abbreviations are written out as they are said (the rules come from kokoro-js, Kokoro's own JavaScript library). eSpeak NG turns each sentence into the sounds of English, and Kokoro speaks them in the voice you chose. A sentence too long for the model in one go is spoken in pieces, split at a comma or between words, so nothing is ever cut off.

It is a computer-made voice. It can mispronounce names, numbers and abbreviations, so listen before you publish. It reads English only for now, it has no pitch setting, and it cannot copy anyone's voice.

• Turn speech back into text with Transcribe audio

Your text stays on your device.

The voice runs in your browser: what you type or open is never sent anywhere. The first time, the voice's files (about 120 MB) are downloaded from KeepSimp itself, never from another site, and your browser keeps them, so it works offline after that. Nothing you make is kept when you leave the page.

• Read what the site keeps