作者位次 #5/8ACL 2026Findings· Findings of ACL
On the Fallacy of Global Token Perplexity in Spoken Language Model Evaluation
Chan-Jan Hsu · Liang-Hsuan Tseng · Yi-Cheng Lin · Yen-Chun Kuo · Ju-Chieh Chou · Kai-Wei Chang · Hung-yi Lee · Carlos Busso
摘要
Generative spoken language models pretrained on large-scale raw audio can continue a speech prompt with appropriate content while preserving attributes like speaker and emotion, serving as foundation models for spoken dialogue. In prior literature, these models are often evaluated using “global t...