<?xml version="1.0" standalone="yes"?>
<Paper uid="H92-1034">
  <Title>Subphonetic Modeling for Speech Recognition</Title>
  <Section position="6" start_page="177" end_page="177" type="concl">
    <SectionTitle>
5 CONCLUSION
</SectionTitle>
    <Paragraph position="0"> In this paper, we developed the framework of senones -state-dependent subphonetic unit. Senones are created by clustering states of triphone models. Thus, we reduced the the number of system parameters with the senone codebook, which renders finer acoustic modeling and provides a way to learn the model topology. In the mean time, we can construct senonic baseforms to improve phonetic baseforms and learn new words without enlarging system parameters.</Paragraph>
    <Paragraph position="1"> Senonic baseforms are constructed by quantizing the states of estimated word models with the senone codebook. We demonstrated that senones can not only significantly improve speaker-independent continuous speech recognition but also have a novel application for new word learning.</Paragraph>
  </Section>
class="xml-element"></Paper>