A crucial step in processing speech audio data for information-extraction, topic detection, or browsing/playback is to segment the input into sentence and topic units. Speech segmentation is challenging, since the cues typically present for segmenting text (headers, paragraphs, punctuation) are absent in spoken language. We investigate the use of prosody (informationgleaned from the timing and melody of speech) for these tasks. Using decision tree and hidden Markov modeling techniques, we combine prosodic cues with word-based approaches, and evaluate performance on two speech corpora, Broadcast News and Switchboard. Results show that the prosodic model alone performs on par with, or better than, word-based statistical language models—for both true and automatically recognized words in news speech. The prosodic model achieves comparable performance with significantly less training data, and requires no hand-labeling of prosodic events. Across tasks and corpora, we obtain a significant improvement over word-only models using a probabilistic combination of prosodic and lexical information. Inspection reveals that the prosodic models capture language-independent boundary indicators described in the literature. Finally, cue usage is task and corpus dependent. For example, pause and pitch features are highly informative for segmenting news speech, whereas pause, duration and word-based cues dominate for natural conversation.

Abstract not found

This thesis is an inquiry into the nature of the high-level, rhetorical structure of unrestricted natural language texts, computational means to enable its derivation, and two applications (in automatic summarization and natural language generation) that follow from the ability to build such structures automatically. The thesis proposes a first-order formalization of the high-level, rhetorical structure of text. The formalization assumes that text can be sequenced into elementary units; that discourse relations hold between textual units of various sizes; that some textual units are more important to the writer's purpose than others; and that trees are a good approximation of the abstract structure of text. The formalization also introduces a linguistically motivated compositionality criterion, which is shown to hold for the text structures that are valid. The thesis proposes, analyzes theoretically, and compares empirically four algorithms for determining the valid text structures of ...

Identifying whether an utterance is a statement, question, greeting, and so forth is integral to effective automatic understanding of natural dialog. Little is known, however, about how such dialog acts (DAs) can be automatically classified in truly natural conversation. This study asks whether current approaches, which use mainly word information, could be improved by adding prosodic information. The study is based on more than 1000 conversations from the Switchboard corpus. DAs were handannotated, and prosodic features (duration, pause, F0, energy, and speaking rate) were automatically extracted for each DA. In training, decision trees based on these features were inferred

We present a domain-independent topic segmentation algorithm for multi-party speech. Our feature-based algorithm combines knowledge about content using a text-based algorithm as a feature and about form using linguistic and acoustic cues about topic shifts extracted from speech. This segmentation algorithm uses automatically induced decision rules to combine the different features. The embedded text-based algorithm builds on lexical cohesion and has performance comparable to state-of-the-art algorithms based on lexical information. A significant error reduction is obtained by combining the two knowledge sources. 1

Abstract not found

This paper addresses the problem that a listener experiences when attempting to capture information presented during a lecture, meeting, or interview. Listeners must divide their attention between the talker and their notetaking activity. We propose a new device -- the Audio Notebook -- for taking notes and interacting with a speech recording. The Audio Notebook is a combination of a digital audio recorder and paper notebook, all in one device. Audio recordings are structured using two techniques: user structuring based on notetaking activity, and acoustic structuring based on a talker's changes in pitch, pausing, and energy. A field study showed that the interaction techniques enabled a range of usage styles, from detailed review to high speed skimming. The study motivated the addition of phrase detection and topic suggestions to improve access to the audio recordings. Through these audio interaction techniques, the Audio Notebook defines a new approach for navigation in the audio domain.

Abstract not found

Centering was formulated as a model of the relationship between attentional state, the form of referring expressions, and the coherence of an utterance within a discourse segment (Grosz, Joshi and Weinstein, 1986; Grosz, Joshi and Weinstein, 1995). In this chapter, I argue that the restriction of centering to operating within a discourse segment should be abandoned in order to integrate centering with a model of global discourse structure. The within-segment restriction causes three problems. The first problem is that centers are often continued over discourse segment boundaries with pronominal referring expressions whose form is identical to those that occur within a discourse segment. The second problem is that recent work has shown that listeners perceive segment boundaries at various levels of granularity. If centering models a universal processing phenomenon, it is implausible that each listener is using a different centering algorithm.The third issue is that even for utterances within a discourse segment, there are strong contrasts between utterances whose adjacent utterance within a segment is hierarchically recent and those whose adjacent utterance within a segment is linearly recent. This chapter argues that these problems can be eliminated by replacing Grosz and Sidner's stack model of attentional state with an alternate model, the cache model. I show how the cache model is easily integrated with the centering algorithm, and provide several types of data from naturally occurring discourses that support the proposed integrated model. Future work should provide additional support for these claims with an examination of a larger corpus of naturally occurring discourses.

In this thesis, we develop computational tools for analyzing conversations based on nonverbal auditory cues. We develop a notion of conversations as being made up of a variety of scenes: in each scene, either one speaker is holding the floor or both are speaking at equal levels. Our goal is to find conversations, find the scenes within them, determine what is happening inside the scenes, and then use the scene structure to characterize entire conversations. We begin by