Unlimited Vocabulary Synthesis Using Line Spectrum Pairs

Abstract

This report describes a new speech synthesis system based on the line spectrum pair (LSP) representation of the speech signal. The system includes a library of approximately 250 stored speech segments, represented by LSP parameters, that are concatenated to produce an unlimited output vocabulary. All components of the system are discussed, and particular emphasis is placed on the methods used to adjust the pitch, duration, and amplitude of concatenated segments to generate natural sounding speech. A new method of computing fundamental frequency contours for sentential intonation patterns is presented. Test results show good segmental intelligibility for the synthesized speech: a score of 86.3 was obtained on the Diagnostic Rhyme Test (DRT); a score of 86.7 was obtained on the Modified Rhyme Test (MRT).

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Sep 30, 1991
Accession Number
ADA241634

Entities

People

  • Stephanie S. Everett

Organizations

  • United States Naval Research Laboratory

Tags

Communities of Interest

  • Autonomy

DTIC Thesaurus Topics

  • Algorithms
  • Amplitude
  • Boundaries
  • Communication Systems
  • Computers
  • Consonants
  • Equations
  • Excitation
  • Frequency
  • Frequency Domain
  • Human-Computer Interaction
  • Human-Computer Interfaces
  • Line Spectra
  • Military Research
  • Resonant Frequency
  • Spectra
  • Standards

Readers

  • Speech Processing/Speech Recognition.

Technology Areas

  • AI & ML
  • AI & ML - Machine Translation