It will likely improve over other speech systems (in general, systems with lower WER trained on variety of conditions/accents/acoustic settings are more robust). We also have Project Euphonia, which specifically targets diverse speech patterns.

→ View original post on X — @jeffdean,