Lip Service: speech and song on your CPU
Giles Hall, November 19, 2026
Why
Lip Service speaks and sings in the same sixteen voices, on an ordinary CPU. The engine is 1.1 MB and the models 56 MB, and it runs faster than real time on one core, in a browser or from Python.
What it does
It reads text aloud in sixteen voices, with numbers, dates and abbreviations said the way a person would say them, and SSML for the rest. It sings from a written score: notes, a syllable under each, and vibrato you draw or borrow from a trained singer. The playground does both, and the editor lets you write a song for up to sixteen singers and hear every change.
Numbers
On one core of an Apple M3 Max it reads 12 times faster than real time on its hifi sound and 23 times on fast.
Licensing
It is free under the GNU AGPL. A closed-source product needs a commercial license.
What's next
It speaks English only today. Bug reports and ideas are welcome: open an issue on GitHub.
Get it
pip install lipservice