Machine learning for multimodal interaction

InhaltsverzeichnisMLMI 2004. Accessing multimodal meeting data encompasses systems, challenges, and opportunities. It includes browsing recorded meetings with Ferret and discusses meeting modeling within multimodal research. The text explores artificial companions and presents Zakim, a software system designed for large-scale teleconferencing, aimed at enhancing computer understanding of human interactions. It introduces a multistream dynamic Bayesian network for meeting segmentation and highlights the use of static documents as structured interfaces to multimedia meeting archives. An integrated framework for managing video collections is discussed, alongside the NITE XML Toolkit's application to the ICSI Meeting Corpus for import, annotation, and browsing. S-SEER focuses on selective perception in multimodal office activity recognition. The content covers mapping speech to images using continuous state space models and introduces an online algorithm for hierarchical phoneme classification. It explores predicting optimal fusion candidates in biometric authentication and presents a mixture of SVMs for face class modeling. The AV16.3 audio-visual corpus is detailed for speaker localization and tracking, along with the 2004 ICSI-SRI-UW meeting recognition system. The adequacy of baseform pronunciations is examined, alongside tandem connectionist feature extraction for conversational speech recognition. Additional topics include s