<?xml version="1.0" standalone="yes"?>
<Paper uid="H93-1020">
  <Title>ON THE USE OF TIED-MIXTURE DISTRIBUTIONS</Title>
  <Section position="8" start_page="106" end_page="106" type="concl">
    <SectionTitle>
6. SUMMARY
</SectionTitle>
    <Paragraph position="0"> This paper provided an overview of work using tied-mixture models for speech recognition. We described the use of tied mixtures in the SSM as well as several innovations in the training algorithm. Experiments comparing performance for different parameter allocation choices using tied-mixtures were presented. The performance of the best tied-mixture SSM is comparable to HMM systems that use similar input features. Finally, we presented a general method we are investigating for modeling segmental dependence with the SSM.</Paragraph>
  </Section>
class="xml-element"></Paper>