Andreas Stolcke

A discriminative adaptation that reads the final prompt token's hidden state through a classification head, producing a label in one forward pass without modifying the backbone, gives a controlled comparison of generative and discriminative inference in an otherwise identical speechLLM.

These findings suggest that turn-taking is primarily conveyed through intonation and silence patterns rather than semantic completeness, enabling faster and more reliable systems without expensive text inference.

Targeted synthetic data can boost multilingual LLM performance by over 6%, while also mitigating catastrophic forgetting on out-of-distribution tasks.