Voice control?

[derUli]

Will this project also make it possible to vocally talk to the duck, similar to Google Assistant or Siri?
As I understand the README, the project currently only supports written commands.
The duck has a microphone, so it should be technically possible to implement speech recognition.

[rokbenko]

Not yet, in either direction: every input today is text (--goal, a .duck file, or MCP chat), and no robot here has text-to-speech yet either.

Voice fits without needing a redesign, though. The Pi Zero 2 W only has 512 MB of RAM total, and this project already treats that as a tight budget: the camera process gets its own memory limit and an OOM score just so a slow client can't take the walk loop down with it. Speech recognition, even a small model, sits in that same category, so keeping it off the board isn't about it being impossible. It's the same headroom problem detection, and the model itself is already avoided by running elsewhere.

So the natural shape is: stream mic audio off the robot, run STT wherever quackd itself already runs (a laptop or a cloud API), and feed the transcript into the exact same free-text goal that --goal and MCP chat already accept. No new contract, no new verb, just a new front door that produces the same string. Talking back needs a second piece on top: an actual TTS voice piped to a robot that currently only has a handful of built-in sounds to draw from.

Nothing like this is implemented or on the roadmap yet. PRs welcome if anyone wants to take a swing at it.