<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-1140">
  <Title>Learning to Say It Well: Reranking Realizations by Predicted Synthesis Quality</Title>
  <Section position="8" start_page="1118" end_page="1118" type="concl">
    <SectionTitle>
5 Conclusions
</SectionTitle>
    <Paragraph position="0"> In this paper, we have presented a method for adapting a language generator to the strengths and weaknesses of a particular synthetic voice by training a discriminative reranker to select paraphrases that are predicted to sound natural when synthesized. In contrast to previous work on this topic, our method can be employed with any speech synthesizer in principle, so long as features derived from the synthesizer's unit selection search can be made available. In a case study with the COMIC dialogue system, we have demonstrated substantial improvements in the naturalness of the resulting synthetic speech, achieving two-thirds of the maximum possible gain, and raising the average rating from &amp;quot;ok&amp;quot; to &amp;quot;good.&amp;quot; We have also shown that in this study, our discriminative method significantly outperforms an approach that performs selection based solely on corpus frequencies together with target and join costs.</Paragraph>
    <Paragraph position="1"> In future work, we intend to verify the results of our cross-validation study in a perception experiment with na&amp;quot;ive subjects. We also plan to investigate whether additional features derived from the synthesizer can better detect unnatural pauses or changes in speech rate, as well as F0 contours that fail to exhibit the targeting accenting pattern.</Paragraph>
    <Paragraph position="2"> Finally, we plan to examine whether gains in quality can be achieved with an off-the-shelf, general purpose voice that are similar to those we have observed using COMIC's limited domain voice.</Paragraph>
  </Section>
class="xml-element"></Paper>