<?xml version="1.0" standalone="yes"?>
<Paper uid="H91-1010">
  <Title>New Results with the Lincoln Tied-Mixture HMM CSR System 1</Title>
  <Section position="8" start_page="66" end_page="68" type="concl">
    <SectionTitle>
DISCUSSION AND CONCLUSIONS
</SectionTitle>
    <Paragraph position="0"> While the additional work on semiphone models has not yielded any improvements over the original semiphone systems, they still represent a potentially useful tradeoff. They still yield a 20-30% higher error rate than do triphone models, but provide more than an order of magnitude reduction in the number of states required in a large vocabulary recognition system.</Paragraph>
    <Paragraph position="1"> The improved duration model, as tested here, is extremely simple way to reduce the error rate by about 10%.</Paragraph>
    <Paragraph position="2"> A better method for determining the minimum state durations might be to perform a Viterbi alignment of the training data and determine the desired splitting factor from the observed minima.</Paragraph>
    <Paragraph position="3"> The new training strategy, while it did not improve performance as tested, did yield results consistent with a method of rapid speaker adaptation. This method of speaker adaptation, which is performed by a modified TM trainer, is well suited to the current DARPA applications.</Paragraph>
    <Paragraph position="4"> The bigram back-off language model was added to the Lincoln CSR. This made the system operational with a more practical class of language models than the previously implemented finite state grammars. In particular, it made testing on the ATIS CSR task feasible.</Paragraph>
    <Paragraph position="5"> The tripling of error rates obtained on the ATIS task compared to the RM task is quite reasonable. A perplexity 25.7 bigram back-off language model trained on 8K RM sentences resulted in an approximate doubling of the error rate compared to the WPG\[12\] and the perplexity 17.8 ATIS bigram language model was trained on only 4K sentences.</Paragraph>
    <Paragraph position="6"> Thus, only a factor of about 1.5 increase occurred due to the extemporaneous speech and the less controlled environment.</Paragraph>
    <Paragraph position="7"> Given the limited time between distribution of the data and the evaluation tests, it has not been possible to adequately study the difficulties unique to the ATIS database nor has it been possible to adequately test our systems.</Paragraph>
    <Paragraph position="8"> There are some known difficulties with the systems reported here (a bug in the recognition network generation has been found) and some known phenomena have not been modeled.</Paragraph>
    <Paragraph position="9"> We tested our best system-to-date and hope to be able to improve the modeling and cure the system difficulties in the near future.</Paragraph>
  </Section>
class="xml-element"></Paper>