• Aucun résultat trouvé

b Perception of musical excerpts

Being moved: literally and metaphorically 1

F.3. b Perception of musical excerpts

Out of the 280 recordings, only 60 were shown to our participants since we focused on the following experimental conditions: deadpan, projected, congruent, and incon-gruent. Consequently, we computed another PCA in order to adapt the organization of the computed features to the representative sample. It resulted in the extraction of four principal components (PC) representing 60% of the total variance. The first two are relatively similar to the first PCA. The first PC was related to spectral

F.3. RESULTS 189

Figure F.1 – Estimated values of each principal component based on the exper-imental conditions and the excerpts. From top to bottom, PC1, PC2, PC3, PC4, and PC5. Within each subgraph, the first bar represents the deadpan condition, the second represents the projected condition, the third represents the congruent condition, and the fourth represents the incongruent condition. All contrasts are FDR-corrected [*: p <0.05,

**: p <0.01, ***: p <0.001].

information and negatively associated to the volume occupied by the violinist. It is labelled PCspect. The second PC encompasses mode and meter while being nega-tively linked to spectral information. It is labelled PCmode. However, the last two PC differed from the first analysis. The third component was associated with the total distance travelled and cepstrum (inverse Fourier transform of the logarithm of the estimated spectrum of a signal) and negatively associated with directness of gestures, while also being negatively associated with mode, roughness, and energy of sound. It is labelledPCcepst. The last component is centered on motion features associated with volume (convex hull and density) and movement (kinetic energy, acceleration, and norm). It is labelled PCmotion.

In the perception phase of this study, we asked our participants to rate both how much they like the excerpt and how expressive it was. These two ratings were highly correlated (r=.77) (Supplementary Material F.5.g). We also asked them to rate the five metaphor categories of the GEMMES. These results are presented in two subsections, respectively.

Liking & Expressive intent

We focused on how different groups would rate the performances based on the four aforementioned experimental conditions: deadpan, projected, congruent, and incon-gruent. In order to account for multiple cofounding factors in our analysis, we used linear mixed models (LMM). These types of model take advantage of modelling random intercept effects such as the random variability induced by each partici-pant, musician, type of excerpt (audio or video), and excerpts. We computed a model encompassing the experimental condition and the musical expertise of our participants (music lover, musician, or violinist) as well as their interaction as fixed effects. We reported a significant improvement in the model when modelling their interaction compared to models encompassing only the main effects for both the evaluation of liking and expressive intent (χ2liking(1, N = 2040) = 21.09,p = 0.002 and χ2expressive(1, N = 2040) = 44.5,p < 0.001, respectively, Figure F.2, full statis-tics Supplementary Material F.5.h). At all levels of expertise, we observed that the deadpan condition is generally perceived as less likeable and less expressive. This difference between deadpan and all the other conditions was significant for the musi-cian and violinist groups (p <0.001, full statistics inSupplementary Material F.5.i).

Music lover showed only one significant difference between the deadpan and congru-ent condition for both rating scales (p < 0.05). Across groups, when focusing on the deadpan condition, we observed a significant difference between the music lover

F.3. RESULTS 191 group and both expert groups (p < 0.01). Finally, the musician group also showed significant decrease in ratings between both the projected and congruent conditions and the incongruent condition (p < 0.01).

Figure F.2 – Estimated ratings for both liking and expressive intent based on the experimental conditions and the level of expertise of the participant. On top, the liking ratings. At the bottom, the expressive intent ratings. All contrasts are FDR-corrected [*: p <0.05, **: p <0.01, ***: p <0.001].

To observe the effect of the performance characteristics, represented by the com-puted PCs, we ran another linear mixed model. While this model encompasses the same random intercept effects, it uses only the PCs as fixed effect to encap-sulate a more general effect across all groups and all conditions. Except for PC-spect, we observed that all PCs have a significant impact on both ratings scales.

For liking, we reported a positive slope between the PCmode, PCcepst, PCmo-tion and the liking scales (χ2P Cmode(1, N = 2040) = 25.32,χ2P Ccepst(1, N = 2040) = 15.34,χ2P Cmotion(1, N = 2040) = 31.01,p < 0.001, Figure F.3, full statistics Sup-plementary Material F.5.j). We report similar effect for the expressive rating scale (χ2P Cmode(1, N = 2040) = 29.02,χ2P Ccepst(1, N = 2040) = 22.72,χ2P Cmotion(1, N =

2040) = 44.97,p < 0.001).

Figure F.3 – Estimated ratings for both liking and expressive intent based on principal component. On top, the liking ratings. At the bottom, the expressive intent ratings. All contrasts are FDR-corrected [*: p <0.05, **: p <0.01, ***: p <0.001].

Metaphorical scales

The second evaluation reported by our participant was related to the metaphors perceived in the performance. First, since every excerpt was selected to represent a particular metaphor, we computed a generalized linear mixed model capturing the interaction between the metaphorical scales and the excerpts. The random intercept effects modelled the variability induced by the participants, the musicians, and the type of excerpt (audio and video). This model outperformed a model without the interaction (χ2(24, N = 2040) = 109.51,p < 0.001, full statistics in Supplementary Material F.5.l). It highlight the fact that each excerpt, across all the experimental in-structions given to the violinists, was evaluated with a specific pattern of metaphors.

Partita andReveuse, for example, showed opposite patterns since they represent op-posite metaphors (Supplementary Material F.5.k). Second, in order to capture the

F.3. RESULTS 193