Hear it
Made with this tool, straight out of the browser, not edited.
Transcript
The old lighthouse keeper climbed the narrow stairs each evening, carrying a lamp and a notebook full of weather. He wrote down the wind, the colour of the sea, and the ships that passed. Nobody ever asked to read the notebook. But on the night of the great storm, it was the only thing that told the rescue boats where the rocks had moved.
Transcript
Is your furnace making that noise again? Summit Heating and Air has same-day repairs, upfront prices, and technicians on call 24/7. Call 555-201-4410 today, or book online at summitheating.com.
Transcript
Here's how it works. Paste your text, pick a voice, and press Generate. A thirty-minute article takes about two minutes on a recent laptop, and you can download the result as an MP3 or a WAV. If a name comes out wrong, like Moxisync, you can tell the voice how to say it.
How to turn text into speech
- Paste your text. Paste it into the box, or press Try a sample. Articles, scripts and whole chapters are fine.
- Pick a voice. Press the play button on any voice to hear it first.
- Set the speed and pauses. Add [pause] or *emphasis* where you want them.
- Press Generate speech. Listen, then download the MP3 or WAV.
What makes it different
Every voice here was checked by a speech recogniser, for pace, for clipping and for background whine, then approved by ear. Robotic or garbled voices were dropped.
The model runs in your browser. Nothing you type leaves your device.
Up to 30,000 characters per render, read in parts with a real progress bar and the time left.
Full-quality audio, free for any use, including commercial.
Pauses, emphasis and names
| Type this | What you hear |
|---|---|
| [pause] | A short pause (0.6 s) |
| [pause 2s] | A pause of the length you give, up to 10 s |
| *really* | Stress on the word |
| [Moxisync](mocks ee sink) | The name said the way you spell it in brackets |
| A blank line | A longer pause between paragraphs |
Privacy
Your text is turned into speech by your browser on this page. It is not uploaded, stored or seen by anyone. The page downloads the voice model once from Hugging Face and the rest from this site, and keeps them in your browser’s cache.
More audio and video tools
- Voicemail greeting generator: a business phone greeting from a template, with 8 kHz files for phone systems.
- Noise remover: take fans, traffic and hum out of a voice recording you made yourself.
- Auto captions: put subtitles on the video you voiced, in your browser.
- Silence remover: cut the pauses and filler words out of a talking video.
- Vocal remover: an instrumental from any song, to put under a voice-over.
- Video compressor: make the finished video small enough for Discord, WhatsApp or email.
Questions
Is this text to speech really free?
Yes. There is no sign-up, no watermark and no daily limit. Each render can be up to 30,000 characters, about half an hour of speech, and you can make as many as you like. It is free because the voice model runs on your own device, so it costs us nothing per use. Pro (coming soon) adds batch files, very long text and more voices.
Is my text uploaded anywhere?
No. The voice model is downloaded to your browser once and runs on your computer or phone. Your text and the audio never leave your device, and once the model is cached the tool also works offline.
How natural do the voices sound?
They use Kokoro-82M, an open model that ranked first among open models on the TTS Arena leaderboard when it was released. We only offer voices that pass our checks: a speech recogniser has to understand every test sentence, the pace has to sound natural, the audio must not clip or carry a background whine, and we listened to every voice ourselves. Some voices did not pass and are not offered.
Can I use the audio commercially?
Yes. You can use what you make in videos, podcasts, ads, courses, phone systems and products, with no credit required. The voices are built into the Kokoro-82M model, which is released under the Apache 2.0 licence; none of them is a copy of a real person’s voice.
How do I add pauses or emphasis?
Type [pause] for a short pause or [pause 2s] for a longer one, put *stars* around a word to stress it, and leave a blank line between paragraphs for a longer break. To fix how a name is said, write it as [Moxisync](mocks ee sink). Pasted SSML with <break> and <emphasis> tags works too.
How long does it take?
On a laptop with a recent browser, a paragraph takes a few seconds, and a long article a minute or two. The first time, the browser downloads the voice model (about 90 MB, or about 325 MB on computers that can use the graphics card), then keeps it for next time. The progress bar shows the time left.
Which formats can I download?
MP3 (small, plays everywhere), WAV (lossless, 24 kHz mono) and, in browsers with an AAC encoder (Chrome, Edge, Safari), M4A. For phone systems, the voicemail greeting maker also exports 8 kHz WAV and μ-law files.
Which languages are supported?
English (US) and Spanish. We only add a language when its voices pass our listening checks; Hindi did not pass yet, and there is no Urdu voice good enough yet.
Is there an Urdu or Hindi voice?
Not yet, and we would rather say so than offer a bad one. The open Hindi voices we tested read too fast and lost sounds, and there is no open Urdu voice good enough, so neither is offered. Subtitles and transcripts in Hindi and Urdu do work: see the caption generator.
Credits and licences
- Kokoro-82M v1.0 by hexgrad, hexgrad/Kokoro-82M, Apache 2.0, in the ONNX build by onnx-community (Apache 2.0). The voices are built into the model; none is cloned from a real person.
- Pronunciation dictionaries from misaki by hexgrad (Apache 2.0). Spanish pronunciation rules are our own.
- ONNX Runtime Web (MIT) runs the model. Audio files are written with Mediabunny (MPL-2.0); M4A uses your browser’s own AAC encoder.
- MP3 files are made with the LAME MP3 encoder 3.100 (LGPL-2.0-or-later, unmodified, as WebAssembly), loaded as a separate file only when you download an MP3: its source, licence and how to replace it (LGPL text).