Attention and the dynamics of tension/release in music
A key concept in the music appreciation is that of the succession of tension and release episodes as time progresses throughout a musical piece. Quantifying or defining precisely what this notion is has been difficult, although it is one that listeners experience reliably (Swain & Swain, 1997). Models that attempt to predict the occurrence and value of these epochs have exploited the structural categories of the notes and chords, and their structural relationships within the musical context (Farbood, 2012). More specifically, tension/release cycles are assumed to reflect musical progressions (or sequences) that are commonly followed in Western music. Listeners are therefore expected to internalize and react to these musical norms that they learn by “passive exposure” to their own music. Consequently, we conjecture that tension/release epochs in the listening experience would induce heightened neural responses that reflect the “tension” cycle. In a recent EEG experiment, we collected EEG signal while listeners were exposed to a recording of Schubert Lied. Using a model based on canonical correlation analysis (CCA) (De Cheveigné et al., 2018), which orders the components corresponding to the best rotations of spectro-temporal filtered signals, we estimated a projected EEG and a projected audio that maximized the correlation between the two signals. We hypothesized that when subjects feel musical tension, they effectively become more attuned to the musical signal, and that in turn is reflected in a significantly enhanced EEG encoding of the music signal (Pelofi et al., 2017). Therefore, tension/release cycles may correspond to increased/decreased levels of correlation between the EEG signal and the audio, which one can objectively compute by a continuous correlation window on the projected EEG and projected audio (see upper panel in Fig. 1). The red curve in lower panel of Figure 1 give credence to this hypothesis, where the critical tension instants predicted by a computational tension model (Farbood et al. 2012) are mirrored by peaks in the EEG-audio correlation function. We believe that such neural markers of the tension/release episodes also exist in ECoG recordings, and discovering them would usher new studies into the objective characterization of the musical experience.