<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-1613">
  <Title>Letter-to-Sound Conversion for Urdu Text-to-Speech System</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Text-to-speech synthesis is logically divided into two stages. The first stage takes raw text input, processes it and converts it into precise phonetic string to be spoken, appropriately annotated with prosodic markers (e.g. stress and intonation). The second stage takes this phonetic representation of speech and generates the appropriate digital signal using a particular synthesis technique. These stages may be referred to as Natural Language Processing (NLP) and Speech Synthesis (SS) respectively (e.g. Dutoit 1997, p.14).</Paragraph>
    <Paragraph position="1"> For SS, formant based techniques (e.g. Klatt 1980) or diphone based techniques (e.g. Dutoit 1997) are normally employed and are generally script independent (as they are only dependent on temporal and spectral acoustic properties of the language and take input in script-neutral form, e.g.</Paragraph>
    <Paragraph position="2"> in IPA). However, NLP is very dependent on cultural and linguistic specific usage of script.</Paragraph>
    <Paragraph position="3"> NLP may also be divided into further parts. The first component is dedicated to pre-processing, 'cleaning' and normalizing input text. Once the input text is normalized, the second component does phonological processing to generate a more precise phonetic string to be spoken. One of the first tasks in the Phonological Processing Component is to convert the input text into a phonemic string using Letter-to-Sound (LTS) rules. This string is then eventually converted to precise phonetic transcription after application of sound change rules and other annotations, as explained later. This paper overviews Urdu writing system, phonemic inventory, NLP for TTS and gives details of the LTS rules for Urdu (also see Rafique et at. (2001) and Hussain (1997: Appendix A), for introductory work).</Paragraph>
  </Section>
class="xml-element"></Paper>