How? Google needs speech where it knows what is said to train the software. That's most easily done with audiobooks or transcripts conversations, not random people speaking without it written down somewhere.