Acoustic structure of the five perceptual dimensions of timbre in orchestral instrument tones

Abstract

Attempts to relate the perceptual dimensions of timbre to quantitative acoustical dimensions have been tenuous, leading to claims that timbre is an emergent property, if measurable at all. Here, a three-pronged analysis shows that the timbre space of sustained instrument tones occupies 5 dimensions and that a specific combination of acoustic properties uniquely determines gestalt perception of timbre. Firstly, multidimensional scaling (MDS) of dissimilarity judgments generated a perceptual timbre space in which 5 dimensions were cross-validated and selected by traditional model comparisons. Secondly, subjects rated tones on semantic scales. A discriminant function analysis (DFA) accounting for variance of these semantic ratings across instruments and between subjects also yielded 5 significant dimensions with similar stimulus ordination. The dimensions of timbre space were then interpreted semantically by rotational and reflectional projection of the MDS solution into two DFA dimensions. Thirdly, to relate this final space to acoustical structure, the perceptual MDS coordinates of each sound were regressed with its joint spectrotemporal modulation power spectrum. Sound structures correlated significantly with distances in perceptual timbre space. Contrary to previous studies, most perceptual timbre dimensions are not the result of purely temporal or spectral features but instead depend on signature spectrotemporal patterns.

Perceptual Learning of Acoustic Noise Generates Memory-Evoked Potentials

“…For audition, perceptual learning has indeed been observed through distributed changes in ''tonotopic'' frequency maps, using pure tones [12,13]. As noise has a flat spectrum on average, a distributed code could not rely on tonotopy, but it could possibly recruit more-complex timbre maps [14,15]. In contrast, a local code posits dramatic changes at the single-neuron or small-network level, with small neural populations expressing the full benefit of learning [7,16].…”

Neural processing of natural sounds

“…Second, in terms of behaviour, it is interesting to note that the fluctuating physical sound characteristics that show the power law characteristics in natural sounds are those that are directly linked to perceptual attributes. Whereas we are unable to perceive the details of the sound pressure waveform, the time varying amplitude yields a percept of intensity fluctuations, rhythm and timbre; the time varying pitch profile carries the melody in a musical phrase; and the spectral envelope contains critical information for other timbral qualities of sound including speech formants (the high-energy frequency bands in voiced human speech that code the identity of vowels and other phonological information) [19]. Finally, these observed natural statistical structures have implications in terms of neural coding.…”

Embodied Listening and Timbre

“…Similar FFT parameters were carried out in MEDS with the addition of a Bartlett window; transients were trimmed above a threshold value of 20, and the resulting calculations were a long-term average centroid across all remaining frames, as described by Kendall and Carterette (1996). Elliott et al (2013) found that perceived timbral noisiness was associated solely with spectral features of timbre; our analysis thus considered only spectral and not temporal or spectrotemporal timbral features. Six spectral parameters were assessed: high-frequency energy, spectral centroid, inharmonicity, spectral flatness, zerocross rate, and auditory roughness, to be discussed in the next section (Table 1).…”

“…Psychophysiological methods may also be useful in order to assess the arousal dimension of timbre without cognitive input. It could also be illuminating to incorporate temporal and spectro-temporal components into an account of timbral noisiness: although steady-state spectrum appears to be of primary importance in evaluations of noisiness and in timbre perception more generally (Elliott et al, 2013;Hajda et al, 1997), time-variant features not addressed here certainly mediate the psychodynamics of listening to noisy timbre as well.…”

“…The musical acoustics literature has identified a broad range of spectral features as comprising the basis for the perception of noisiness in musical timbre (Elliott, Hamilton, & Theunissen, 2013), including inharmonicity, or spectral energy outside the harmonic series (Moore, Peters, & Peters, 1985); aperiodic components such as sidebands, broadband noise, and subharmonics (Gibiat & Castellengo, 2000); combination tones; and auditory roughness (Helmholtz, 1877(Helmholtz, /1954Sethares, 1998). Such acoustic accounts have informed numerous case studies: for example, Fales and McAdams (1994) explored inharmonic components in the timbral makeup of African instruments and their perception as either ''layered'' or ''fused'' with the harmonic content of the tone.…”