A voice for our ads with open-source text-to-speech

Our ads need a voice. We use Kokoro, an open-source text-to-speech model.

Why not record a voice?

We will one day. But for a small project, a model that we can run again and again is practical. When the script changes, the voice changes in a minute.

Runs locally

The model is small enough to run on a normal CPU, inside a Docker container. No online service and no cost per word.

Short, plain sentences

Text-to-speech sounds best with short, clear sentences. That fits our writing style anyway.

The licence

The model and its voices have an open licence that allows commercial use. We check the licence of everything that we use.