New AI Speech Coach Cut Public Speaking Anxiety by a Third

Most AI speech coaches today work like a blunt instrument. They count your "ums," measure your pace, and hand you the same tips they'd give anyone else. The problem: a nervous student defending a thesis and a manager pitching to a client need very different help. So a team of researchers built a system that takes your speech's context into account — the topic and situation — before it judges how you did. It then generates an improved version of your own speech, in both text and audio, so you can hear what better sounds like.

The results, published for a European engineering education conference, came from a small pilot: seven undergraduate students over one to three weeks. Each practiced the same speech at least three times. Every single participant improved. Filler word usage went down, and self-reported anxiety dropped by 15.3% to 34.4% — meaning the most anxious participants got roughly a third calmer. Students said the tailored feedback and the rewritten example speeches made them feel more confident, not just more polished.

The catch is size. Seven people over a few weeks is a hint, not proof. There's no control group taught the old-fashioned way, so some of the gain could simply come from practicing repeatedly — which, to be fair, is itself a lesson. The system also only listens to your voice. It can't see your posture, your pacing around the stage, or whether you're staring at your shoes.

That's the researchers' own next step: adding video analysis of physical behavior. If it works at scale, this kind of tool could sit in job-prep courses, university communication classes, or your phone before a big meeting. You wouldn't need a coach, a camera crew, or a room full of strangers to get honest, personalized feedback.