Brenna now talks: Aaron Aldana’s site assistant gets an open-source voice
The assistant on aaronaldana.com now speaks with Kokoro, an open-source voice model running on his own server instead of a paid voice service.
Brenna, the AI assistant at the bottom of aaronaldana.com, now answers out loud in a natural voice — and the voice runs on an open-source model on Aaron Aldana's own server, not a paid voice service.
Why it matters
Natural-sounding speech used to mean paying a voice provider for every sentence, which gets expensive for an assistant that talks to every visitor. Open-source speech models have closed most of that gap. Brenna is a small, real example of a business swapping a metered service for a model it runs itself.
How it works
- The model: Kokoro, an 82-million-parameter open-source text-to-speech model, released under the Apache 2.0 licence.
- The voice: one of Kokoro's built-in voices, chosen by Aldana for sounding closest to a real person.
- Where it runs: a small service of its own, with the model built into it, so there is no per-sentence bill and no outside service holding the recordings.
- Speed: replies are spoken a sentence at a time, so Brenna starts talking before the whole answer has been turned into audio.
- Fallback: if the voice service is ever unavailable, Brenna falls back to the browser's built-in voice instead of going silent.
What Brenna does
Brenna answers questions about Aldana's work — Creai, Arbi, Activate and the rest — and can take a project brief and turn it into a proposal request, so a visitor can go from a question to a conversation about working together without filling in a form.
Try it at aaronaldana.com. More from Activate Global Media on Aaron Aldana.
