<?xml version="1.0" standalone="yes"?>
<Paper uid="H90-1061">
  <Title>IMPLEMENTATION ASPECTS OF LARGE VOCABULARY RECOGNITION BASED ON INTRAWORD AND INTERWORD PHONETIC UNITS</Title>
  <Section position="12" start_page="317" end_page="317" type="concl">
    <SectionTitle>
8. CONCLUSIONS
</SectionTitle>
    <Paragraph position="0"> This paper provided a detailed presentation of all aspects of the implementation of a large vocabulary speaker independent continuous speech recognizer to be used as a tool for the development of recognition algorithms based on hidden Markov models and Viterbi decoding. The complexity of HMM recognizers is greatly increased by the introduction of detailed context dependent units for representing interword coarticulafion. A vectorized representation of the data structures involved in the decoding process, along with compilation of the connection information among temporally consecutive words, has led to a speed up of the algorithm of about one order of magnitude. An average recognition time of about one minute per sentence (on the computer configuration used in the experiments), although far from real time, allows us to perform a series of training experiments and to tune the recogmtion system parameters in order to obtain high performance recognition on complex tasks such as the DARPA resource management.</Paragraph>
  </Section>
class="xml-element"></Paper>