The role of static spectral properties in vowel identification
James M. Hillenbrand, Brendan J. McMahon · The Journal of the Acoustical Society of America · 1987
The classic study of Peterson and Barney [J. Acoust. Soc. Am. 24, 175–184 (1952)] demonstrated that vowels in CVC context are identified with a very high degree of accuracy by human listeners. A number of attempts have been made to classify the Peterson and Barney vowels using some combination of F0 and the three lowest formant frequencies. Although these attempts have been reasonably successful, the accuracy of the classification algorithms has been well below that of human listeners. Human listeners, however, have access to dynamic (duration and spectral change) as well as static information (F0 and “target” formant frequencies). The purpose of the present study was to determine the identifiability of the Peterson and Barney vowel set based exclusively on static information. A formant synthesizer was used to generate isolated, steady-state versions of all 1520 vowels (76 speakers × ten vowels × two repetitions) in the Peterson and Barney data base using measured values of F0 and F1–F3. Preliminary results from a small group of trained listeners suggest that the error rate for the resynthesized stimuli is 3.5–4.5 times greater than the error rate reported in the original Peterson and Barney study. Pilot data also suggest that the identification task may be extremely difficult for some untrained listeners. [Work supported by Rome Air Development Center under contract ♯ F3060285-C-0008.]