https://github.com/skorokithakis/havpe-server
It'll just send the recognized speech to your own API endpoint, and speak whatever is returned.