Learns from millions of hours of diverse human speech.

Performance varies across regional dialects and non-native speakers.

Users often must speak commands like "comma" or "period."