Midam Kim begins with a voice-agent call that repeatedly mishandled her spelled name, requested an ambiguous account identifier and interrupted her response. She separates these failures into recognition, pronunciation, vocabulary, turn-taking and conversational-understanding problems rather than treating them as one generic model error.
Midam Kim presents listening and speaking as complementary channels organized around sounds, words, interaction and mental models. These layers are interdependent: good recognition alone cannot compensate for poor timing or failure to understand what the user is trying to accomplish.
Midam Kim explains why the user's evolving mental model matters when spoken history is not visibly retained as it is in text chat. She highlights clarification, emotion handling, context retention and dynamic orchestration, then asks whether voice systems can adapt as speakers and language change. The framework is diagnostic rather than a one-step fix.
Watch on YouTube
