Text to Speech
Turn text into speech in Spanish or English with AI that runs in your browser. Pick a voice and speed, listen, and download the audio as WAV. Free, and the text is never sent.
What this tool does
Type or paste some text, pick a voice and a speed, and an artificial intelligence model reads it out loud. You can listen to the result on the page and download it as WAV to use in a video, a presentation or a voice message.
The model is Kokoro 82M, an open voice model (Apache 2.0 license) that runs inside your browser. The text is never uploaded to any server: no account, no monthly character quota, no watermark on the audio.
How to use it
- Type or paste the text (up to 3,000 characters).
- Pick a voice: three in Spanish (Dora, Alex and Santa) and five in English, from the US and the UK.
- Adjust the speed if you want it read slower or faster.
- Press Generate audio. The first time, the model is downloaded; you will see its size before it starts and a progress bar.
- Listen to the result and press Download WAV.
Long texts are split into sentences, each one is generated and they are joined into a single audio with a short pause between them.
System voices: the no-download option
In the System voices tab the text is read with the voices your operating system already has (Windows, macOS, Android or iOS). They download nothing and play instantly, but they have two limits: the browser does not let you save them as a file, and they differ from one device to another. They are fine for quickly hearing a text; to download the audio, use the AI voice.
Privacy: the text never leaves your device
Typical text-to-speech services send your text to a server that generates the audio and sends it back. Here everything happens in your browser: the only thing that travels over the network is the model download, the first time. You can convert an unpublished script, an internal document or a personal message without anyone else reading it.
Requirements and download size
- Lightweight version (CPU): about 111 MB — the quantized Kokoro model (92 MB) plus the eSpeak NG phonemizer (18 MB), which turns text into phonemes.
- GPU version (WebGPU): about 344 MB. It is faster; the tool offers it only if your browser supports WebGPU, and you can turn it off to get the lightweight one.
Everything is stored in the browser cache, so next time it starts instantly. It works in recent desktop browsers and on phones with enough memory; on a modest phone it may be slow or run out of memory with long texts.
What to know about the voices
Kokoro was trained on far more English audio than Spanish, and it shows: the English voices sound more natural. The Spanish ones are clear and fine for narration, but they can sound somewhat flat or stress an uncommon word oddly.
Spanish text is turned into phonemes with eSpeak’s Latin American variant, so «c» and «z» sound like «s» and «ll» like «y». Numbers are read as words. For acronyms and unusual proper names, spelling them the way they are pronounced usually gives a better result.
Frequently asked questions
Is it free? Is there a limit?
It is free and needs no account. The only cap is 3,000 characters per conversion, because your own device does the work and a much longer text would take several minutes. For long texts, convert them in parts.
Is my text sent to a server?
No. The voice model is downloaded to your browser and the conversion happens right there. The only thing that travels over the network is the model download, the first time; the text you type never leaves your device.
How much is downloaded?
About 111 MB the first time: the lightweight Kokoro 82M model (92 MB) and the eSpeak NG phonemizer (18 MB). If your browser supports WebGPU you can use the faster GPU version, which weighs about 344 MB. Everything stays in the browser cache and is not downloaded again.
What accent do the Spanish voices have?
Kokoro's three Spanish voices (Dora, Alex and Santa) sound like neutral Spanish. Text is turned into phonemes with eSpeak's Latin American variant, so «c» and «z» are pronounced as «s» and «ll» as «y», as in Mexico. They are less polished than the English voices: the model was trained on far more English audio.
What is the difference between the AI voice and the system voices?
System voices are the ones your operating system ships with: they play instantly and download nothing, but the browser does not let you save them as a file and they differ from one device to another. The AI voice is generated in your browser, sounds the same on any device and can be downloaded as WAV.
Can I use the audio in my videos or projects?
Kokoro 82M is released under the Apache 2.0 license, which allows commercial use. The audio you generate is yours to use; still, check the rules on AI-generated content of the platform where you publish it.
Why is it slow?
The first time, because of the download. After that it depends on your device: on CPU a short sentence takes a few seconds and a text of several paragraphs can take a minute or two. With the GPU (WebGPU) it is considerably faster.
Related tools
Others from the catalogue that pair well with this one.