Another Thursday, another Google Meet Up to talk about Google Assistant.
We are making progress:
We've learned that trying to get Google and robot talking over Wi-Fi doesn't work well with all of us on a Google Meet also talking over Wi-Fi.
We've learned that splitting up a sentence and looking for familiar words is better than getting it to process one word by itself.
The cleverest thing we did was say all of our command words over and over, looked for Google's common mistakes, and added those in as duplicate command words:
if text == "left" or
text == "let" or
text == "lift" or
text == "laughed" or
text == "lit":
s.sendall(b'goLeft')
if text == "right" or
# etc
The biggest issue still to overcome is Google's stoppy-starty conversational nature. The code that came with Google AIY assumes you'll be using it like a normal human being, asking it to to your maths homework or play something by Danny Gonzales.
With that in mind, Google seems to listen in chunks, and then stops listening to process whatever it heard or didn't hear. This means that for every five seconds it listens, there's a second or two where it isn't listening, and that's annoying.
We want it to always be listening, and whenever it hears a word it likes, it sends a command. We will keep working on it.
Comments