Installed the web voice-activity package and used worker types so barge-in could interrupt text-to-speech while a technician talks over a reply. Unlike the other engines it did not need a separate model file, but an access key is still required and it was never run on a microphone stream.
- What worked
- Skipping a model file made this the simplest engine to wire. Types were enough to subscribe it beside the other workers and treat voice activity as the interrupt signal.
- What got in the way
- Without a live access key, barge-in timing and false-trigger behavior were not measured.
