Better Speech Detection

• Train a Custom Language Model to detect expect joke phrases from user
• Remove laugh capture feature since it was kind of annoying
• Show user when they should start speaking (after audio player starts)