Automatic detection of speaker state: Lexical, prosodic, and phonetic approaches to level-of-interest and intoxication classification

摘要: Traditional studies of speaker state focus primarily upon one-stage classification techniques using standard acoustic features. In this article, we investigate multiple novel features and approaches to two recent tasks in detection: level-of-interest (LOI) detection intoxication detection. the task LOI prediction, propose a Discriminative TFIDF feature capture important lexical information Prosodic Event approach AuToBI; combine these with for new multilevel multistream prediction feedback similarity-based hierarchical fusion learning approach. Our experimental results outperform published all systems 2010 Interspeech Paralinguistic Challenge - Affect Subchallenge. task, evaluate performance Event-based, phone duration-based, phonotactic, phonetic-spectral based approaches, finding that combination phonotactic achieve significant improvement over 2011 Speaker State Intoxication Subchallenge baseline. We discuss our their implications future research.