Speech Synthesis in Festival

A practical course on making computers talk

Edition 1.4.1, for Festival Version 2.0

last updated May 8th 2000

NOTE: this document is incomplete

• 1 Introduction
  • 1.1 Text to Speech
• 2 History
• 3 Festival Overview
  • 3.1 Users of Festival
  • 3.2 Core system
  • 3.3 Festival 1.4.1
  • 3.4 Festival uses
  • 3.5 Using Festival
  • 3.6 Exercises
  • 3.7 Hints
• 4 Architecture
  • 4.1 Overview
  • 4.2 Utterance architectures
  • 4.3 Festival utterance structure
    • 4.3.1 Utterances, relations and items
    • 4.3.2 Standard relations
    • 4.3.3 Items and features
  • 4.4 Synthesis modules
    • 4.4.1 Utterance types and modules
  • 4.5 Exercises
  • 4.6 Hints
• 5 Text processing
  • 5.1 Text analysis
  • 5.2 Identifying tokens
  • 5.3 Chunking into utterances
  • 5.4 Tokens to words
  • 5.5 Summary of Festival's text processing
  • 5.6 Text modes
  • 5.7 An example email text mode
  • 5.8 Mark-up languages
  • 5.9 Normalization of Non-Standard Words
  • 5.10 Two interesting text analysis problems
    • 5.10.1 Tokenization without whitespace
    • 5.10.2 Number pronunciation
  • 5.11 Exercises
  • 5.12 Hints
• 6 Linguistic/Prosodic processing
  • 6.1 Lexicons
    • 6.1.1 letter to sound rules
  • 6.2 Part of speech tagging
  • 6.3 Prosodic phrasing
    • 6.3.1 Phrasing by decision tree
    • 6.3.2 Phrasing by statistical models
    • 6.3.3 Measuring phrasing algorithms
  • 6.4 Intonation
    • 6.4.1 Accent assignment
    • 6.4.2 F0 generation
    • 6.4.3 Some intonation models
    • 6.4.4 F0 Extraction
  • 6.5 Duration
  • 6.6 Post-lexical rules
  • 6.7 Summary of Linguistic/Prosodic processing
  • 6.8 Exercises
  • 6.9 Hints
• 7 Waveform synthesis
  • 7.1 Diphones
    • 7.1.1 Diphone schema
    • 7.1.2 Recording diphones
    • 7.1.3 Labelling diphones
  • 7.2 Getting longer Units
    • 7.2.1 Finding the best unit
    • 7.2.2 Acoustic cost
    • 7.2.3 Hunt and Black Target Cost
      • 7.2.3.1 Estimating the acoustic cost
      • 7.2.3.2 Hunt and Black weight training
    • 7.2.4 Black and Taylor Unit Clustering
    • 7.2.5 Taylor and Black: Phonological Structure Matching
    • 7.2.6 Continuity cost
    • 7.2.7 Selecting the units
  • 7.3 Joining the units
  • 7.4 Exercises
  • 7.5 Hints
• 8 Building models from databases
  • 8.1 Collecting databases
  • 8.2 Labelling databases
  • 8.3 Extracting features
  • 8.4 Building models
  • 8.5 Exercises
  • 8.6 Hints
• 9 New Voices
  • 9.1 Building New Voices
  • 9.2 Limited domain synthesis
  • 9.3 Voice Conversion
• 10 Future directions
  • 10.1 Synthesis of any voice
  • 10.2 Size of database
  • 10.3 Autolabelling
  • 10.4 Training from labels
• 11 Final comments
• 12 References

This document was generated on 30 May 2000 using the texi2html translator version 1.52.