<?xml version="1.0" standalone="yes"?>
<Paper uid="H91-1012">
  <Title>BYBLOS SPEECH RECOGNITION BENCHMARK RESULTS</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
INTRODUCTION
</SectionTitle>
    <Paragraph position="0"> This paper will present new test results for running the BBN BYBLOS system on the speech recognition benchmarks in both the Resource Management (RM) and the Air Travel Information System (ATIS) domains.</Paragraph>
    <Paragraph position="1"> During this reporting period we have conceentrated on speaker-independent recognition conditions. However, we will also report a new result demonstrating the need and usefulness of speaker adaptation in order to be able to recognize the speech of speakers with different dialects than those found in the training data.</Paragraph>
    <Paragraph position="2"> For the RM corpus, we report on three conditions:  1. The common SI-109 training condition that has been widely reported in the past.</Paragraph>
    <Paragraph position="3"> 2. The new SI-12 training paradigm that we introduced at the previous DARPA workshop.</Paragraph>
    <Paragraph position="4"> 3. Adaptation to the dialect of the speaker  The ATIS domain presents a new type of speech recognition problem in several respects. First of all, and most importantly, the speech was collected during simulations of actual use of the ATIS system. The speakers were completely uncoached, and therefore, the range of speech phenomena goes far beyond that of the carefully controlled read-speech conditions that exist in the RM corpus. We will describe our recent efforts to deal with these new problems.</Paragraph>
    <Paragraph position="5"> Since understanding is the ultimate goal of the ATIS domain, we use a rank ordered list of the N-best speech recognition hypotheses as the interface to the natural language component of BBN's spoken language system. Below, we desoribe a new procedure which allows the system to use powerful but eomputationally prohibitive acoustic models and statistical grammars to reorder the hypotheses in the N-best lisL For the ATIS corpus, we report on two conditions:  1. A baseline control condition using a standard training set, lexicon, and M-gram grammar.</Paragraph>
    <Paragraph position="6"> 2. An augmented condition using additional training, acoustic mod null els for nun-speech phenomena, and a 4-gram class grammar. In the next section, we describe the main features of the baseline Byblos system used in both RM and ATIS tests. Next, the RM results are presented. For the ATIS domain, we first describe the speech corpus used. Then we describe the informal baseline training condition which was developed to provide informative controlled experiments for this domain. Next, we explain how the Byblos system was modified for this evaluation. Finally, we describe our augmented condition and present comparative results.</Paragraph>
  </Section>
class="xml-element"></Paper>