About this plugin
Codeep Voice for Posts turns a post into an audio file. You open the plugin’s
screen in your admin, pick a post and press a button. Your own browser generates
the speech, and the finished recording is saved to your media library as an
ordinary MP3, attached to that post. A small player then appears on the post for
your readers.
The speech is produced by a model that runs on your computer, inside the admin
screen. The plugin has no server of its own. There is no account to create, no
API key to paste and nothing that counts your usage, and your post text is not
uploaded anywhere.
Hear it before you install
Two recordings of the same sentence, one made by this plugin and one by the
speech voice built into your browser, are on the plugin page:
io/codeep-voice-for-posts/
What you get
- A real MP3 file in your media library, attached to the post
- Nine voices: Heart, Bella, Nicole and Sarah; Michael and Adam, American; Emma and George, British; and Amy, American, fast on a computer with no graphics card
- A player on the post, under 12 KB, which takes its colours from your theme
- A download button in that player, so a reader can keep the file
- Your readers download the MP3 and nothing else: no model, no engine, no waiting
Because the file is a plain MP3 in your media library, you can do anything you
would do with any other upload: download it, put it in a podcast feed, serve it
from a CDN, or hand it to someone.
How it works
The voice model is downloaded once and then lives in your browser. When you press the button, your own computer generates the audio, the plugin saves it as an MP3 attachment, and your visitors simply get that file.
Visitors download nothing but the MP3 itself. No model, no engine, no WebGPU on their side.
Honest about speed
The voice is generated on your own computer, so the speed is your computer’s speed, and it depends on how long the post is.
On a machine with graphics acceleration it is quick. Without acceleration it still works, it just takes considerably longer on a long article.
The plugin measures your computer and shows an estimate for that specific post before you start, so you know what you are in for. Either way you only do it once per post, and the file stays for good.
You can also find out before installing anything. This page downloads the same
voice and times one sentence on your own machine:
io/check/
What this plugin does not claim
It does not make your site compliant with any accessibility standard. WCAG does not require that article text be read aloud, so an audio version is a convenience for your readers rather than a compliance measure. A good convenience, but a convenience.
External services
This plugin downloads a text-to-speech model the first time you use it. Nothing is sent out: the transfer is one-way, into your browser.
- Hugging Face ( huggingface.co ) hosts the Kokoro-82M model files, licensed Apache-2.0. Downloaded once, when you press Prepare the voice, and then cached in your browser. Your post text is never uploaded. Terms: https://huggingface.co/terms-of-service Privacy: https://huggingface.co/privacy
- Hugging Face ( huggingface.co ) also hosts the Piper voice files, licensed MIT, used by the voice named “Amy, American, fast on any computer”. The same one-way download applies: it happens once, when you press Prepare the voice with that voice selected, and the file is then cached in your browser. Which file is downloaded depends on the voice you choose, and its size is shown on the screen before you start. Terms: https://huggingface.co/terms-of-service Privacy: https://huggingface.co/privacy
No account, API key or registration is required for any of this, and no data about you or your visitors is sent anywhere.
Third-party libraries
The plugin carries the code it needs instead of pulling it from a CDN. That is
what makes it work offline, and it means shipping other people’s work. Licences
are included in assets/js/vendor/licenses/.
- kokoro-js 1.2.1, Apache-2.0: the voice model runner
- @huggingface/transformers 3.8.1, Apache-2.0: inside the same bundle
- onnxruntime-web 1.22, MIT: the WebAssembly runtime in assets/wasm/,
js and part of the same bundle
- phonemizer 1.2.1, Apache-2.0: turns text into the sounds a voice reads,
js and inside the same bundle
- espeak-ng, GPL-3.0-or-later: the pronunciation rules phonemizer is built
on. It ships inside phonemizer rather than as a package of its own
- @breezystack/lamejs 1.2.7, LGPL-3.0: MP3 encoding
Two of those licences are the reason this plugin is GPLv2 or later rather
0. Both are
compatible through the “or later” clause, and neither would be without it.
espeak-ng is easy to miss, so it is named here on purpose: it appears in no
json. It arrives inside phonemizer, which arrives inside kokoro-js,
and all three end up compiled into one file.
0) is not bundled. It is
downloaded from Hugging Face on first use, as described above.