Comparing the fundamental frequencies of resolved and unresolved harmonics: Evidence for two pitch mechanisms?

Abstract

Four experiments measured sensitivity (d′) to differences in fundamental frequency (F0) between two simultaneously presented groups of frequency-modulated harmonics. Each group was passed through a bandpass filter in either a LOW (125–625 Hz), MID (1375–1875 Hz), or HIGH (3900–5400 Hz) frequency region. In the first two experiments, a dynamic F0 difference (ΔF0) was created by introducing a 180° disparity between the frequency modulations imposed on the two groups. Experiment 1 measured sensitivity to such ΔF0’s between a MID group with a baseline F0 of 125 Hz and all components summed in sine phase, and a HIGH group, in four conditions. When the baseline F0 of the HIGH group was also 125 Hz, performance was good when its components were summed in sine phase and bad when they were in alternating phase. Conversely, when the HIGH F0 was 62.5 Hz, performance was better for alternating phase than for sine phase, consistent with alternating phase doubling the internal representation of HIGH group’s F0. Similar results were obtained for a comparison between the LOW and MID groups. Experiment 2 measured sensitivity to ΔF0’s between either the LOW and MID groups or between the MID and HIGH groups, for baseline F0s of 88 and 250 Hz. Sensitivity was best when the combination of frequency region and F0 was such that both groups were resolved or both unresolved by the peripheral auditory system, and worst when the groups differed in ‘‘resolvability.’’ Experiment 3 replicated experiment 2 using a different paradigm, in which the two groups were always modulated coherently, and in which the ΔF0 was constant throughout the signal. Experiments 2 and 3 also measured sensitivity to differences in F0 between successively presented tokens of the same group. Experiment 4 showed that the high sensitivity to (simultaneous) ΔF0’s when both groups were unresolved could be attributed to listeners detecting the pitch pulse asynchronies that inevitably arise from ΔF0’s. Finally, a method for predicting sensitivity to simultaneous, across-frequency ΔF0’s from that to successive within-channel differences, on the basis of optimum combination of information, was applied to the results of experiment 3. The method succeeded in predicting sensitivity to ΔF0’s between two groups of resolved harmonics, but over-estimated performance when one group was resolved and the other unresolved. The results suggest that the F0’s of resolved and unresolved harmonics are processed by separate mechanisms, in contrast to the predictions of ‘‘autocorrelation’’ models of F0 encoding.

Responses of cochlear nucleus neurons to harmonic and mistuned complex tones

“…Two vowels with identical f0 that differ in timbre can be identified with accuracy greater than chance (Assmann and Summerfield 1989), but accuracy improves dramatically if the f0s of the individual vowels differ by as little as 3% (Assmann and Summerfield 1990;Bregman 1989, 1993;de Cheveigné 1993de Cheveigné , 1997de Cheveigné , 1999. Pairs of non-speech harmonic tones can also be separated, presumably by similar processes; however, these sounds have been studied less often (Beerends and Houtsma 1989;Carlyon and Shackleton 1994;Carlyon 1996;Micheyl et al 2006). …”

Evidence for two pitch encoding mechanisms using a selective auditory training paradigm

“…The harmonics were all added in sine (0°) phase and were bandpass-filtered digitally with the use of a filter with a flat top and 48 dB/octave slopes; components to which an attenuation larger than 48 dB would have to be applied were omitted. Following Shackleton and Carlyon (1994) and Carlyon and Shackleton (1994), we used two nominal F0s (88 and 250 Hz) and three filtering regions (a low region with lower and upper corner frequencies of 125 and 625 Hz, a mid region of 1375-1875 Hz, and a high region of 3900-5400 Hz), leading to six different F0-region combinations. According to Shackleton and Carlyon and Carlyon and Shackleton, the harmonics of both 88 and 250 Hz are resolved in the low region and unresolved in the high region; in the mid region, the resolution status of the harmonics depends on the F0 used: They are unresolved at 88 Hz but resolved at 250 Hz.…”

On Ideal Binary Mask As the Computational Goal of Auditory Scene Analysis

“…It is well-known that the auditory system can resolve the first few harmonics, while higher harmonics are unresolved. Psychoacoustic research suggests that the auditory system may use different mechanisms to deal with resolved and unresolved harmonics (Carlyon and Shackleton, 1994;Bird and Darwin, 1997). Subsequently, Hu and Wang (2003;) developed a CASA model that employs different mechanisms in the low-and the high-frequency range.…”