下記のとおり、AI So-Go-Chi 研究会を開催いたします。
本研究会では、Travis Ralph-Donaldson を講師にお迎えし、ハイブリッド形式で開催いたします。ご興味のある方はぜひご参加ください。また、関心をお持ちの方々へのご周知もお願いいたします。

日時：2026 Feb. 20 (Friday) 13:00 -14:00
開催形式： ハイブリッド開催

• 会場： RIEC main building M431
• オンライン (Zoom)：
  　ミーティングID：977 9538 1904
  　パスコード：028627

講演者: Travis Ralph-Donaldson
Founder and CEO of Niter (foundational AI startup)
PhD Candidate at the University of Southampton

題目:
“Beyond Audio: Multimodal Speech AI from Research to Real-World Impact”

概要
Speech recognition systems remain predominantly audio-based, leaving them vulnerable to noisy, real-world conditions and limiting their potential across education and healthcare. This talk presents a journey from PhD research to commercial deployment, exploring how multimodal approaches — combining audio with 3D facial depth data and visual speech cues — can fundamentally improve how machines understand human speech. I will discuss recent work on silent motion interpretation and lip-based evaluation submitted to IEEE FG 2026 (SMILE), which investigates the comparative value of 2D video and 3D facial mesh data for phoneme-level recognition under varying noise conditions. I will then show how insights from this foundational research have been translated into two applied domains: Speacher, an AI-powered pronunciation coaching application currently deployed in 40 UK schools and the subject of a recent linguistics journal submission examining the role of phonological decoding in foreign language acquisition; and the Talk More Tracker, a speech therapy monitoring project aimed at improving outcomes for stroke recovery patients. Finally, I will introduce Niter, a UK-based AI startup building a foundational multimodal speech recognition model designed for on-device inference, and discuss the path from academic research to commercialisation. The talk will conclude with reflections on UK–Japan collaboration opportunities in speech AI.