Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore the technical challenges behind voice AI interruptions in this 27-minute conference talk that examines why ChatGPT's Advanced Voice Mode and other voice AI agents struggle with natural turn-taking in conversations. Learn how interruptions represent an unsolved problem across all voice AI systems, causing user frustration and immediate call abandonment when agents cut users short or misinterpret sounds like coughs as conversation cues. Discover how human conversation serves as a framework for understanding current turn detection approaches and future developments in the field. Gain insights into linguistic perspectives on turn detection in human dialogue, identify what's working and what's broken in existing methods, and understand potential solutions for building voice AI systems that interrupt users less frequently than human family members. The presentation covers the various forms of turn-taking failures, from premature agent interruptions to false positive detection of user speech, and examines why people have significantly less patience for conversational mistakes from machines compared to humans.
