EAM and music perception

There is an interesting discussion in the online journal Empirical musicology review (EMR) concerning the use of electroacoustic music (EAM) for the study of music perception. In "Time series analysis as a method to examine acoustical influences on real-time perception of music" (EMR, vol. 5, no. 4 (October, 2010)), Roger T. Dean and Freya Bailes use an extract from Trevor Wishart's Red bird to analyse correlations between the acoustic properties intensity and spectral flatness, and listener arousal (perceptions of change) and valence (expressed affect). They argue that

"[p]revious studies of listeners’ real-time perceptions of affect in music have attempted to map response through time to acoustic properties of the piece [...]. Missing are substantial attempts to assess which acoustic properties also drive listeners’ perceptions of the structure of the same music. Structure in this instance need not be a music-theoretic analysis of large-scale form [...], but refers to the low-level assessment by a listener of change and continuity in the music. [... M]usical forms that do not rely on hierarchical structures such as tonality or meter might exhibit quite a close relationship between acoustic properties of the work, listener perceptions of structure (change in sound), and listener perceptions of affect. EAM is one such form, and the subject of the current paper."

Dean and Bailes find that

"intensity influences perceptions of change and expressed arousal substantially. Spectral flatness influences valence, while animate sounds influence the valence response and its variance."

Marcus T. Pearce, in "Time-series analysis of Music: Perceptual and Information Dynamics" (EMR, vol. 6, no. 2 (April, 2011)), comments that Dean and Bailes

"[...] give two reasons for using EAM in their study: first, to demonstrate that their methods generalise beyond Western tonal music which is more often used in empirical work on music perception; and second, Red Bird provides an opportunity to test their methods on idiosyncratic temporally-localised timbral features in addition to the continuous features which generalise to other musical genres (see, e.g., Dean, Bailes & Schubert, 2011). Interestingly, their timbre feature of choice is spectral flatness, which they view as a more global indicator of timbre than spectral centroid, which is more commonly used in research on music perception (though this is not true of research on audio signal processing and music information retrieval where spectral flatness is one of the standard descriptors used in the MPEG 7 standard)."

Spectral flatness is the geometric mean of the power spectrum divided by the arithmetic mean. Noisy spectra have high flatness, peaky spectra low flatness. Spectral flatness is also related to the information content of the sound. The spectral centroid, i.e. the mean, barycenter or "mass center" of a spectrum, is correlated with brightness. Both flatness and centroid are included in the MPEG 7 standard. For descriptions of these and other timbre measures, see Geoffroy Peeters, A large set of audio features for sound description, 2004.

In addition, Pearce remarks that

"Dean and Bailes also argue that EAM can be algorithmically generated in such a way that the acoustic and algorithmic parameters of interest are systematically varied in creating stimuli for research on music perception. In other work, for example, Dean et al. (2011) extend their approach to the effects of intensity on arousal in two pieces written by Roger Dean, one of which is composed in the minimalist style. We might legitimately ask what advantage such algorithmically generated music has over the stimuli often constructed artificially to create experimental conditions in empirical research on music perception. The most obvious advantage is that the results gain in ecological validity from using stimuli created by composers, using stylistically legitimate methods, with an artistic purpose. These results should generalise to the experience of similar music outside the laboratory, while results obtained with artificially created or altered musical stimuli are not guaranteed to do so. The advantage of computer-generated music over other musical styles is that it can be produced so as to conserve experimental control."

Dean and Bailes respond to Pearce in another paper, "Modelling perception of structure and affect in music: spectral centroid and Wishart’s Red Bird" (EMR, vol. 6, no. 2 (April, 2011)) where they analyse the Red Bird extract using spectral centroid and find that

"[...] it is fairly clear that spectral centroid and spectral flatness bear a quite distant relationship to atomic perceptual processes, and it is still unclear how they may influence cognition. But acoustic intensity, on the other hand, is an immediate determinant of an important perceptual response, loudness, and this relationship is much better understood. Again, most studies use short tones, often synthetic, but it is clear that even with longer musical extracts, intensity is a close determinant of continuously perceived loudness."