If you wanted to become a better speaker you should first be able to measure how much attention others are paying to you when you speak. You aren't doing well if people get bored, look away, easily interrupt you, and do other things while you speak.
When training your speaking skills you would wear a cap that has multiple wide-lense pinhole cameras built-in. With them, AI would detect the facial expressions and body language of those around you. It would generate a score for each person you speak to and keep statistics on how much attention they are paying to you when you speak.
During the review, your weaknesses that the AI identified would be shown to you on a playback. They would be linked to relevant sections in the learning material so that you can learn how to improve.
In advanced mode, the AI could give you cues and tips in real time as it detects disinterest in the people you are speaking to.