<?xml version="1.0" standalone="yes"?>
<Paper uid="H92-1082">
  <Title>An Efficient A* Stack Decoder Algorithm for Continuous Speech Recognition with a Stochastic Language Model*</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
INTRODUCTION
</SectionTitle>
    <Paragraph position="0"> Speech recognition may be treated as a tree network search problem. As one proceeds from the root toward the leaves, the branches leaving each junction represent the set of words which may be appended to the current partial sentence. Each of the branches leaving a junction has a probability and each word has a likelihood of being produced by the observed acoustic data. The recognition problem is to identify the most likely path (word sequence, W*) from the root (beginning of the sentence) to a leaf (end of the sentence) taking into account the junction probabilities (the stochastic language model, p(W)) and the acoustic match (including time alignment, p(OIW)) given that path \[2\]:</Paragraph>
    <Paragraph position="2"> where O is the acoustic observation sequence and W is a word sequence.</Paragraph>
    <Paragraph position="3"> This paper is concerned with the network search problem and therefore correct recognition is defined as outputting the most likely sentence W* given the language model, the acoustic models, and the observed acoustic data. If the most likely sentence is not the one spoken, it is a modeling error--not a search error. This paper *This work was sponsored by the Defense Advanced Research Projects Agency. The views expressed are those of the author and do not reflect the official policy or position of the U.S. Government. will assume for simplicity that an isolated sentence is the object to be recognized. (The algorithm extends trivially to recognize continuous input.)</Paragraph>
  </Section>
class="xml-element"></Paper>