Michael D. Hall
Dept. of Speech and Hear. Sci. W1-10, Univ. of Washington, Seattle, WA 98195
Richard E. Pastore
Dept. of Psychol., State Univ. of New York at Binghamton, Binghamton, NY 13902-6000
While several dimensions contribute to voice and instrument timbre, perceptual interactions between dimensions have not been adequately assessed. As a result, the information used in timbre identification is unclear. Three experiments sought to define steady-state timbral features. First, Garner classification reflected the perceptual integrality of spectral slope and formant structure (a source and filter property, respectively); classification speed along either dimension depended upon variability along the other dimension. Feature detection then was evaluated within arrays of distractor pitches with homogeneous timbre. When targets had /a/ vowel formants with shallow spectral slopes, search time increased as a log function of array size consistent with parallel processing. Parallel search was not obtained for targets with /i/ formants and steep slopes, suggesting a feature coded as the presence or absence of /a/ formants with shallow slopes. A final search experiment using heterogeneous distractor timbres showed that this feature was separable from pitch. Search times for conjunctions of pitch and timbre linearly increased with array size, implying serial processing. Thus attention was needed to conjoin attributes to perceive tonal objects. Implications are discussed for the nature of attention and the perception of properties relevant to communication. [Work supported by AFOSR.]