<?xml version="1.0" standalone="yes"?>
<Paper uid="W98-1436">
  <Title>ROMVOX- EXPERIMENTS REGARDING UNRESTRICTED TEXT- TO-SPEECH SYNTHESIS FOR THE ROMANLAN LANGUAGE</Title>
  <Section position="4" start_page="306" end_page="306" type="concl">
    <SectionTitle>
4. Conclusions
</SectionTitle>
    <Paragraph position="0"> * As ~resented before, the aim of our research was to develop an improved synthesis technique that should assure a better quality of the generated signal. The improvement concerns the signal processing part and it presents the following aspects and advantages with respect to our previous developments, respectively to other synthesis techniques.</Paragraph>
    <Paragraph position="1"> The TD-PSOLA (Time Domain Pitch Synchronous Overlap-add) developed by CNET is a very simple but ingenious method which assures high voice quality, the only disadvantage is that it is based on a time: domain windowing technique which can introduce some spectral distortions during the pitch modification. The result of these spectrum distortions can be interpreted as a reverberation of the desired pitch-modified signal. TD-PSOLA requires at the same time a very exact pitch synchronous framing; any  framing error may cause the unpleasant increase of this reverberation effect. The first disadvantage was solved by CNET through adopting the LP-PSOLA technique.</Paragraph>
    <Paragraph position="2"> Our approach doesn't use any windowing technique, so this source of spectral distortion is eliminated. Figure 3. presents the spectral behavior of a generated signal with k=0.66 fundamental frequency modification (decreasing fundamental frequency)respectively with k=l.5 (increasing fundamental frequency), both cases in comparison with the spectrum of the initial signal. As both figures show, the peaks of the modified harmonics are situated almost on the ideal imaginary spectrum envelope.</Paragraph>
  </Section>
class="xml-element"></Paper>