Table 1.
Selected drum samples used as stimuli for the vocal imitations.
Descriptions and articulations are taken from the sample library documentation and are not exhaustive desciptions of the recording setup, strike style or drum machine settings etc. Unless specified otherwise the acoustic drums were struck with a stick.
Fig 1.
Graphical interface of a single test page used for the online listening test.
Listeners were asked to rate the similarity between the imitation (reference) and 6 test items (same–category drum sounds), on a continuous scale from ‘less similar’ to ‘more similar’.
Fig 2.
Contingency tables of the highest rated sound for each imitated (i.e. target) sound, by drum category.
Cell values and shading indicate the proportion (0–1) of tests for a given imitated sound where the rated sound was considered most similar to the imitation. Asterisks in the diagonals indicate cases where the imitated sound was rated most similar to the imitation, significantly above chance (padj <0.05). (A) Cymbals, (B) Hats, (C) Kicks, (D) Snares, (E) Toms.
Fig 3.
Comparison of similarity ratings between imitations and target vs. non–target sounds, by drum category.
Values are mean rating parameter estimates with 95% Wald confidence intervals.
Table 2.
Comparison of similarity ratings between imitations and target vs. non–target sounds, by drum sound.
Values are mean rating parameter estimates with 95% Wald confidence intervals (CIs). Cases where the CIs overlap between conditions for each drum sound are given in bold.
Fig 4.
Comparison of similarity ratings between imitations and target vs. non–target sounds, by imitator.
Values are mean ratings with 95% Wald confidence intervals.
Table 3.
Performance for each of the feature sets, in terms of i) the LMER model fit (AIC), and ii) percentage of imitated drum sounds for which the rating ∼ distance slope is significantly less than 0 (α < 0.05).
Fig 5.
Slope estimates for the LMER model fitted using the distance measure from method PHG.
A negative slope indicates a decrease in perceptual similarity with an increase in distance, i.e. sounds for which the method performs well. Values are mean estimates across all imitations for each drum sound, with 95% Wald confidence intervals.