Effects of fundamental frequency removal and low-pass filtering on pitch comparisons.

Abstract

Pitch has been defined as perceptual attribute which allows the ordering of sounds on a frequency-related scale ordered from low to high (Klapuri & Davy, 2006; p. 8). A primary acoustic dimension for determining the pitch of complex periodic sounds is fundamental frequency (Stevens & Volkmann, 1940; Idson & Massaro, 1978). Relationships between the spectral components of complex tones also contribute to pitch. Harmonic relations, in particular, strengthen the sensation of pitch (Plomp, 1976; Terhardt, 1974).In contrast, timbre traditionally has been defined by exclusion. For example, the American Standards Association (1960) defines timbre as attribute of auditory sensation in terms of which a listener can judge that two sounds similarly presented and having the same loudness and pitch are dissimilar. Part of the reason for the adherence to the traditional definition of timbre may be that timbre has repeatedly been demonstrated to consist of several perceptual dimensions. Yet, establishing an inclusive definition critically depends on understanding the nature of each of the various dimensions of timbre. Research that has sought to define timbre dimensions has typically involved the multidimensional scaling (MDS) of timbre dissimilarity judgments for musical instrument tones. Reported MDS solutions have tended to converge on several acoustic correlates for the perceptual dimensions of timbre. These associations are summarized below (Caclin, McAdams, Smith, & Winsberg, 2005; Grey, 1977; Krumhansl, 1989; McAdams, Winsberg, Donnadieu, De Soete, & Krimphoff, 1995; Miller & Carterette, 1975). The correspondence between some timbre dimensions and their putative acoustic correlates also has been confirmed by timbre discrimination tasks in response to acoustic manipulations (Grey & Moorer, 1977; also see McAdams, Beauchamp, & Meneguzzi, 1999).Dimensions of musical instrument timbre fall into three broad categories: temporal, spectrotemporal, and spectral.1 Important temporal information is conveyed during the attack portion of a tone, the most salient characteristic of which is the logarithm of rise time. For example, MDS studies have shown that the logarithm of rise time is strongly correlated with a perceptual dimension of timbre (Caclin et al., 2005; Krimphoff, McAdams, & Winsberg, 1994; McAdams et al., 1995). When attack transients are removed (Saldanha & Corso, 1964) or swapped across instru- ments (Thayer, 1974), the accuracy of instrument recognition suffers.The most commonly referenced spectrotemporal dimension is spectral flux, a measure of (the lack of) harmonic stability over the duration of a tone (e.g., see Krumhansl, 1989). However, this dimension has not always significantly correlated with MDS solutions, and sometimes corresponding perceptual dimensions have been accounted for equally well or better by spectral attributes such as spectral irregularity (Krimphoff et ah, 1994) or the relative intensity of odd versus even harmonics (Caclin et al., 2005), both of which contribute to the jaggedness of the spectrum. Several alternative spectrotemporal attributes have been determined to contribute to timbre when controlling for spectral flux, as well as spectral envelope shape and rise time (McAdams et ah, 1999; also see Beauchamp & Lakatos, 2002).The current investigation sought to understand the potential impact on pitch made by spectral timbre information. One primary spectral attribute of timbre is brightness, which is related to the relative amount of higher-frequency energy, and is strongly correlated with spectral centroid, an intensity-weighted average of the frequencies in a tone (Schubert & Wolfe, 2006). Because spectral centroid strongly correlates with a primary perceptual dimension in MDS solutions, it has repeatedly been inferred that listeners use perceived brightness to differentiate between musical instruments (Ehresman & Wessel, 1978). …