Final Report on Contract Number N00014-76-C-0483,
Abstract
Synthetic 'set' - 'sat' and (G AH G) - (G AA G) continua were constructed by the simultaneous manipulation of vowel duration and the first formant target frequency of the vowel. In a randomized identification test, a high trained phonetician transcribed perceived vowel quality when instructed to ignore irrelevant changes in vowel duration. Results are compared with a group of native English listeners who made phonemic judgments when presented with the same tape. It was found that there were large individual differences in the phonetic labels assigned to particular stimuli, so that cross-subject comparisons are not easily made. However, the one phonetically trained observer did show evidence of criterial shifts when only duration was changed, calling into question the commonly held notion that phonetically trained observers can estimate vowel quality independently of duration. A second study of factors contributing to the perception of a natural voice quality in synthetic computer-generated speech was initiated. An additive harmonic speech synthesizer has been designed and implemented in software in order to carry out an evaluation of the hypotheses concerning the perceptual importance of various factors contributing to naturalness in synthetic speech.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 1979
- Accession Number
- ADA085829
Entities
People
- Dennis H. Klatt
- June E. Shoup