<?xml version="1.0" standalone="yes"?>
<Paper uid="N06-2006">
  <Title>Class Model Adaptation for Speech Summarisation</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Techniques for automatically summarising written text have been actively investigated in the field of natural language processing, and more recently new techniques have been developed for speech summarisation (Kikuchi et al., 2003). However it is still very hard to obtain good quality summaries.</Paragraph>
    <Paragraph position="1"> Moreover, recognition accuracy is still around 30% on spontaneous speech tasks, in contrast to speech read from text such as broadcast news. Spontaneous speech is characterised by disfluencies, repetitions, repairs, and fillers, all of which make recognition and consequently speech summarisation more difficult (Zechner, 2002). In a previous study (Chatain et al., 2006), linguistic model (LiM) adaptation using different types of word models has proved useful in order to improve summary quality. However sparsity of the data available for adaptation makes it difficult to obtain reliable estimates of word n-gram probabilities. In speech recognition, class models are often used in such cases to improve model robustness. In this paper we extend the work previously done on adapting the linguistic model of the speech summariser by investigating class models.</Paragraph>
    <Paragraph position="2"> We also use a wider variety of objective evaluation metrics to corroborate results.</Paragraph>
  </Section>
class="xml-element"></Paper>