SYNTHETIC SPEECH STUDY.

Abstract

The report describes a technique to automatically evaluate the intelligibility of speech transmitted over a communication channel. The technique is called CORODIM (Correlation Of the Recognition Of Degradation with Intelligibility Measurements). It transmits a test signal composed of speech-like sounds representative of phoneme consonants, and measures, by means of spectral channel analysis, the degradation suffered by each of the test signal constituents. The degradation manifests itself as an 'effective noise spectrum' which is measured and matched to one of a library of reference noise spectra. By means of the spectrum matching operation and a measurement of signal-to-noise ratio each constituent sound of the test signal is assigned a probability of recognition. These values are weighted by phoneme probability of occurrence factors, summed, and normalized to obtain a score representative of word intelligibility based on either initial or final consonant recognition of CVC-type words. CORODIM evaluates scores for both initial and final consonants and takes their product for the overall word intelligibility score. The part of the CORODIM technique following the spectral analysis operation was computer simulated. The technique was checked against word articulation scores under identical speech link conditions. Two types of speech degradations were considered. The first was additive noise having the same spectral characteristics as that used in deriving the library of phoneme recognition probability data. The second type was degradation produced under laboratory simulated speech link conditions. (Author)

Document Details

Document Type
Technical Report
Publication Date
Jun 30, 1966
Accession Number
AD0639964

Entities

People

  • G. Strohmeyer
  • J. Bogusz
  • Roger K. Smith

Tags

DTIC Thesaurus Topics

  • Additives (Chemicals)
  • Communication Channels
  • Computers
  • Consonants
  • Degradation
  • Intelligibility
  • Measurement
  • Phonemes
  • Probability
  • Recognition
  • Spectra
  • Speech

Fields of Study

  • Engineering

Readers

  • Regression Analysis.
  • Speech Processing/Speech Recognition.