If every smart speaker streamed audio to the cloud 24/7, waiting for servers to detect "Hey Alexa." At 16 kHz, that's 77 GB per device monthly; before transcription costs, that can hit thousands of dollars per device. Network latency, privacy concerns from constant audio streaming, and the infrastructure needed to support millions of devices is not feasible.
This is a companion discussion topic for the original entry at https://www.edgeimpulse.com/blog/voice-command-recognition-on-rubik-pi-3-getting-started-guide