Acoustic Correlates of Meaning Structure in Conversational Speech

摘要

We are interested in the problem of extracting meaning struc tures from spoken utterances in human communication. In Spo ken Language Understanding (SLU) systems, parsing of mean ing structures is carried over the word hypotheses generated by the Automatic Speech Recognizer (ASR). This approach suf fers from high word error rates and ad-hoc conceptual repre sentations. In contrast, in this paper we aim at discovering meaning components from direct measurements of acoustic and non-verbal linguistic features. The meaning structures are taken from the frame semantics model proposed in FrameNet, a con sistent and extendable semantic structure resource covering a large set of domains. We give a quantitative analysis of mean ing structures in terms of speech features across human-human dialogs from the manually annotated LUNA corpus. We show that the acoustic correlations between pitch, formant trajecto ries, intensity and harmonicity and meaning features are statis tically significant over the whole corpus as well as relevant in classifying the target words evoked by a semantic frame.