Professional Documents
Culture Documents
Aca 2022 1125
Aca 2022 1125
The Violin but not the Human Voice is the Most Beautiful Timbre
--Manuscript Draft--
Full Title: The Violin but not the Human Voice is the Most Beautiful Timbre
Powered by Editorial Manager® and ProduXion Manager® from Aries Systems Corporation
Masked Manuscript without Author Information Click here to access/download;Masked Manuscript without
Author Information;Masked Manuscript.docx
1 Abstract
2 Beauty is central to aesthetic experiences of music and plays an important role in music
3 choice. Existing studies in empirical aesthetics have mainly focused on the relations
4 between musical features such as pitch and duration and beauty judgments. However, little
5 is known about the relation between musical timbre and beauty judgments. Some 17th- and
6 18th-century thinkers have proposed that the most beautiful timbre is the human voice,
7 followed by the violin and the flute. To test these views, here we asked 44 university music
8 and 41 nonmusic majors to listen to 40 unfamiliar melodic excerpts with each sung by the
9 voice (lu) and played on a violin and a flute and to rate the degree of musical beauty using a
10 7-point Likert item (1 = not at all beautiful to 7 = very beautiful). Bayes factors revealed
11 strong evidence that both music and nonmusic majors gave the highest beauty ratings for
12 the violin version, the lowest ratings for the singing voice version and intermediate ratings
13 for the flute version. These results provide the first experimental evidence against the view
14 that the human voice is the most beautiful, but for the view that the violin is more beautiful
15 than the flute. Our study has potentially significant implications for individuals who create
18
VIOLIN IS THE MOST BEAUTIFUL 2
19 Introduction
22
23 All art attempts to create beauty (Hanslick, 1986). Beauty is the core part of the
24 aesthetic experiences of music (Giannouli et al., 2022; Istók et al., 2009) and an important
25 criterion for music choice (Juslin & Isaksson, 2014). Although it is difficult to define
26 (Nieminen et al., 2011; Sisti et al., 2021), a Likert-type item has been widely used to
27 evaluate the perception of musical beauty (e.g., Anglada-Tort et al., 2019; Juslin et al.,
28 2022). So, what music is beautiful? Much of the prior work has focused on the influences of
29 musical tonality (Brattico et al., 2020; Hur, 2020; Nieminen et al., 2011; Suzuki et al., 2008),
30 harmony (Sarasso et al., 2019; Suzuki et al., 2008), rhythm (Brattico et al., 2020), tempo
31 (Brattico et al., 2020; de Jong et al., 1973; Hur, 2020), and complexity (Marin et al., 2016)
32 on perceived beauty, little is known about the influence of musical timbre on beauty ratings.
34 distinguish two sounds that have the same pitch, duration, loudness, and spatial location
35 (American National Standards, 2013). It is considered the most important basic attribute of
36 music (Seashore, 1936; van Elferen, 2020). Thus, to search for beautiful music, we should
37 understand the relationship between timbre and beauty judgments. The present study
39 It has been believed for a long time that timbre is a significant factor in our
40 appreciation of music’s beauty. In the 17th and 18th centuries, some thinkers such as Doni
41 (1640, as cited in Cypess, 2010), Ancelet (1757, as cited in Dolan, 2013), and Rousseau
VIOLIN IS THE MOST BEAUTIFUL 3
42 (1765; as cited in Vélez, 2021) held the views that the human voice is the most beautiful
43 timbre, and the violin is the most beautiful of all instruments. In other words, these thinkers
44 argued that the human voice is considered to be the most beautiful timbre, followed by the
45 violin, and finally the other instruments such as flute. According to Rousseau (1765), the
46 violin is more beautiful than the flute because the former combines softness with brightness
48 Although it has been assumed that timbre plays a role in the perception of beauty in
49 music, only two studies have investigated this relationship. To examine the effects of vocal
50 and instrumental timbres on musical beauty, Brattico et al. (2011) asked participants to
51 listen to their familiar happy and sad music excerpts with and without lyrics (18 s each) and
52 evaluate the perceived beauty of each excerpt on a 5-point Likert item ranging from 1 (ugly)
53 to 5 (beautiful). They found that instrumental music was rated as more beautiful than vocal
54 music, regardless of the emotional tone of the music. The acoustic feature analysis showed
55 that instrumental music had a lower spectral centroid than vocal music; however, there was
56 no difference in attack slope between the two musics. A study by Gotlieb and Konečni
57 (1985) investigated the effects of two different instrumental timbres on musical beauty. The
58 participants rated the beauty of a piano or harpsichord recording of the Goldberg Variations
59 by Bach. Results showed that the piano version of the piece received higher beauty ratings
60 than the harpsichord version. However, no study to date has directly assessed whether the
61 human voice is considered to be the most beautiful timbre, followed by the violin and the
62 flute.
63 The aim of this study was to directly test the ideas that the most beautiful sound is
64 produced by the human voice, followed by the violin and the flute. To this end, we required
VIOLIN IS THE MOST BEAUTIFUL 4
65 participants to rate the perceived beauty after being exposed to melodies in three different
66 timbres—singing voice, violin, and flute. To replicate previous findings related to music
67 expertise (Jaśkiewicz et al., 2016; Müller et al., 2009, 2010; Nieminen et al., 2011), we
68 included music and nonmusic majors. In order to avoid the influences of familiarity (Marin
69 et al., 2016; Vuoskoski et al., 2022) and music preference (Eerola & Vuoskoski, 2011;
70 Nieminen et al., 2011; Vuoskoski & Eerola, 2017; Vuoskoski et al., 2022) on musical
71 beauty, unfamiliar music was used. This was based on the large correlation between
72 familiarity and preference (North & Hargreaves, 1995; Rawlings & Leow, 2008; Yoo et al.,
73 2018). Given the lack of related research or theory, we were not able to make any specific
74 hypotheses or predictions.
75 Method
76 Subjects
77 Sample size was determined by a priori power analysis using MorePower 6.0
78 (Campbell & Thompson, 2012). To detect a moderate effect size of η2p = .0588 (Cohen,
79 1988) for the main effect of timbre (singing voice, violin and flute) or the interaction effect
80 of timbre and group (music and nonmusic) in a repeated measures analysis of variance
81 (ANOVA) with 80% power at the two-tailed 0.05 significance level, the suggested
82 minimum sample size was 80. Thus, 85 undergraduate or postgraduate students were
83 recruited from two universities. Local ethical committee approval was obtained, and all
84 subjects gave informed consent and were paid for their participation.
85 The sample consisted of 44 music students (22 male and 22 female) who received
86 formal music training and 41 nonmusic students (20 male and 21 female) who received little
87 or no formal music training (see Table 1 for demographic characteristics). All except two
VIOLIN IS THE MOST BEAUTIFUL 5
89 Handedness Inventory (Oldfield, 1971) and reported normal hearing and vision. As shown
90 in Table 1, there was moderate evidence that the two groups were matched in terms of age,
91 handedness and education years. However, there was strong evidence that the music group
92 had more years of music playing and/or singing than the nonmusic group.
93 --------------------------------------------------------
95 --------------------------------------------------------
96 Materials
98 Zhang et al. (2019). The melodies were selected from unknown European operas in the
99 classical and romantic periods to rule out the mere exposure effect and music preference.
100 Half of the melodies expressed happiness and the other half sadness. Three renditions of
101 each melody were performed by a violinist, a flutist, or a vocalist with the syllable “lu” in
102 the bel canto style, resulting in 120 excerpts (see the supplemental materials). All
103 performers were female and had received 18 years of music training. They were asked to
104 perform the pieces in a roughly similar way. These melodies were recorded in mono with a
105 sampling rate of 22050 Hz, 16-bit resolution, and 353-kbps bit rate, and then were
106 normalized to −3 dB with a 1-s fade-out, and were saved in .wav format using Adobe
108 A pretest was conducted in additional 8 music students to test the similarities in
109 performance variations or levels among different timbre versions of each excerpt. The
VIOLIN IS THE MOST BEAUTIFUL 6
110 students rated three timbre versions for their congruency on a 7-point Likert item (1 = very
111 incongruent, 4 = not sure, 7 = very congruent) as for rubato, intensity, phrasing, and the
112 overall performance level. The results showed that variations in timing (M = 5.08, SD =
113 0.22), dynamics (M = 5.14, SD = 0.36), phrasing (M = 5.76, SD = 0.46) and the
114 performance level (M = 5.29, SD = 0.37) were similar across the versions.
115 To derive participants’ beauty ratings from timbral features, we extracted several
116 timbre descriptors from each melody. On the basis of previous studies, we included a
117 brightness-related descriptor—spectral centroid (Saitis & Siedenburg, 2020; Schubert &
118 Wolfe, 2006; Springer et al., 2021) and harshness-related descriptors (Rozé et al., 2017)
119 such as harmonic spectral variation, MFCCratio, and attack slope (see Table 2). Spectral
120 centroid is the geometric center of the spectral frequency (Giannakopoulos & Pikrakis, 2014;
121 McAdams, 1999) and correlates positively with perceived brightness (Saitis & Siedenburg,
122 2020; Schubert & Wolfe, 2006; Springer et al., 2021). Harmonic spectral variation is the
123 average over the sound duration of the one’s complement of the normalized correlation
124 between harmonic peak amplitudes of two adjacent frames (Kim et al., 2006; Peeters et al.,
125 2000) and can positively predict perceived harshness (Rozé et al., 2017). MFCCratio is a
126 ratio between the first two mel-frequency cepstral (MFCC) coefficients (Rozé et al., 2017,
127 2020) and can positively predict perceived harshness (Rozé et al., 2017). Attack slope is the
128 averaged temporal slope of the energy envelope attack during the attack portion of the
129 sound (Eerola et al., 2012; McAdams et al., 2017) and can negatively predict perceived
130 harshness (Rozé et al., 2017). Harmonic spectral variation was extracted using MPEG-7
131 Audio Encoder (Kim et al., 2006); attack slope and other features were extracted using the
132 Timbre Toolbox R2021a (Peeters et al., 2011) and the MIRtoolbox 1.8.1 (Lartillot et al.,
VIOLIN IS THE MOST BEAUTIFUL 7
134 --------------------------------------------------------
136 --------------------------------------------------------
137 Procedure
138 Subjects were tested individually in a quiet room. E-Prime (Version 1.1) running on
139 a laptop was used for stimulus presentation and response collection. First, the subjects were
140 given instructions about rating the musical excerpts. Next, subjects were provided with four
141 practice trials to familiarize them with the task and were asked to adjust the sound volume
142 to their most comfortable listening level. Then, the formal experiment began. Each trial
143 started with a 0.5 s presentation of a gray fixation cross in the center of the black screen.
144 After that, a melody was played through the headphones while a green fixation cross
145 appeared at the middle of the black screen. After listening to the melody, subjects rated how
146 beautiful the music sounded on a 7-point Likert item (1 = not at all beautiful, 4 = neutral, 7
147 = very beautiful) by pressing one of the number keys 1–7 on the computer keyboard. Finally,
148 subjects were asked to press the space bar to initiate the next trial. The melodies were
149 presented pseudorandomly such that melodies with the same type of timbres or emotions
150 did not occur more than three times consecutively. The stimulus order was kept constant
151 across subjects. In order to determine the effectiveness of the musical stimuli, the subjects
152 were also required to rate perceived valence (1 = very unpleasant, 4 = neutral, 7 = very
153 pleasant), perceived arousal (1 = very calm, 4 = neutral, 7 = very exciting), and familiarity
VIOLIN IS THE MOST BEAUTIFUL 8
154 (1 = very unfamiliar, 4 = neutral, 7 = very familiar) of the music excerpts. These ratings
158 Parametric tests often assume normality, homoscedasticity or sphericity (Field, 2017;
159 Mair & Wilcox, 2020; Wilcox, 2017). Violations of these assumptions can result in high
160 probability of type I error, relatively poor power, or inaccurate confidence intervals (Wilcox,
161 2022; Wilcox et al., 2018; Wilcox & Serang, 2017). Thus, the data were first tested for
162 normality with the Shapiro-Wilk test, homoscedasticity with Levene’s test, or sphericity
163 with Mauchly’s test. If the assumptions of analysis of variance (ANOVA) were met, then
164 parametric tests were performed using JASP (Version 0.16.4); if any assumption was
165 violated, then nonparametric tests were conducted (Field, 2017). The aligned rank transform
166 (ART) can be used to perform nonparametric ANOVA (Elkin et al., 2021; Wobbrock et al.,
167 2011). In brief, the ART procedure first aligns the data separately for each effect and then
168 applies midranks to the aligned data. After that, a parametric ANOVA is run on the aligned
169 ranks for every effect (Feys, 2016; Wobbrock et al., 2011; Wobbrock & Kay, 2016). When
170 an interaction or a main effect emerges, nonparametric post hoc multiple comparisons with
171 the ART-C algorithm are followed up (Elkin et al., 2021). Specifically, the ARTool.exe
172 (Version 2.1.2), an open-source Windows application, was first used to produce the aligned-
173 and-ranked data (Wobbrock et al., 2011). JASP was then used to conduct all ANOVAs based
174 on Type III sums of squares and pairwise comparisons with Holm-corrected t tests.
175 To measure the magnitude of the effect, an effect size and its confidence interval (CI)
176 were reported: for the ANOVAs, we used the MOTE package (Version 1.0.2) in R (Version
VIOLIN IS THE MOST BEAUTIFUL 9
177 4.2.1) and RStudio (Version 2022.07.2+576) to calculate ωp2 and 95% CI; for the t tests, we
180 But even so, in the frequentist approach, a nonsignificant result cannot provide
181 evidence for the null hypothesis (Brydges & Bielak, 2020; Keysers et al., 2020; Lakens,
182 2017; Rouder et al., 2009) inasmuch as the experiment might have had relatively low power
183 to detect the true null effect (Cohen, 1994; Harms & Lakens, 2018; Lakens, 2017). One
184 method to evaluate the absence or presence of an effect is to calculate a Bayes factor
185 (Anderson & Maxwell, 2016; Lakens et al., 2020). The Bayes factor (BF) is the ratio of the
186 likelihood of one hypothesis over another and can measure the relative strength of evidence
187 for the alternative (H1) and null (H0) hypotheses (Love et al., 2019; Masson, 2011; Rouder
189 The summary statistics from the F and t tests were used to compute BFs. For the
190 ANOVAs, we used the R package anovaBFcalc (Version 0.1.0) to calculate exact BFs from
191 the values of F and df, with a default Pearson Type VI prior of α = −0.5 (Faulkenberry, 2019,
192 2021). For the subsequent paired t tests, we used the R package BayesFactor (Version
193 0.9.12-4.4) to calculate the BFs from the values of t and n, with a default Cauchy prior of
194 0.707. According to Jeffreys (1961) and Lee and Wagenmakers (2014), 1 < BF < 3 is
195 considered weak evidence for H1, 3 < BF < 10 moderate, and BF >10 strong; 0.33 < BF < 1
196 indicates weak evidence for H0, 0.10 < BF < 0.33 moderate, and BF < 0.10 strong. For
197 instance, a BF of 20 means that the data are 20 times more likely under H1 than under H0,
198 providing strong evidence for the presence of an effect. All raw data and code are available
200 Results
202 To verify the validity of music, a 3 (timbre) × 2 (emotion) repeated measures ANOVA
203 was conducted on valence, arousal, and familiarity ratings. Descriptive statistics are
204 presented in Table 3. Because of deviation from normality, Ws ≤ 0.99, ps ≤ .671, and
205 sphericity, Ws(2) ≤ 0.96, χ2s(2) ≥ 3.79, ps ≤ .151, the data were analyzed with
206 nonparametric ANOVAs. Results are provided in the Table 4. For valence ratings, there was
207 strong evidence for the absence of the main effect of timbre, but strong evidence for the
208 presence of the main effect of emotion and the interaction of timbre and emotion. Post hoc
209 tests showed strong evidence to support that happy music was rated as more pleasant than
210 sad music across the violin, t(84) = 21.44, p < .001, d = 3.03, 95% CI [2.22, 3.84], BF =
211 3.10 × 1032, flute, t(84) = 19.22, p < .001, d = 2.71, 95% CI [1.96, 3.46], BF = 1.69 × 1029,
212 and singing voice versions, t(84) = 16.78, p < .001, d = 2.37, 95% CI [1.68, 3.05], BF =
214 --------------------------------------------------------
216 --------------------------------------------------------
217
218 --------------------------------------------------------
220 --------------------------------------------------------
VIOLIN IS THE MOST BEAUTIFUL 11
221 For arousal ratings, there was strong evidence for the main effects of timbre and
222 emotion and the interaction between timbre and emotion. Post hoc tests provided strong
223 evidence to support that happy music was rated as more exciting than sad music across the
224 violin, t(84) = 20.42, p < .001, d = 2.79, 95% CI [2.03, 3.54], BF = 1.05 × 1031, flute, t(84)
225 = 17.14, p < .001, d = 2.34, 95% CI [1.67, 3.01], BF = 8.85 × 1025, and singing voice
226 versions, t(84) = 15.75, p < .001, d = 2.15, 95% CI [1.51, 2.79], BF = 4.26 × 1023.
227 For familiarity ratings, there was also strong evidence for the main effects of timbre
228 and emotion and the interaction between them. Post hoc tests provided strong evidence to
229 support that happy music was rated as more familiar than sad music across the violin, t(84)
230 = 9.94, p < .001, d = 0.58, 95% CI [0.36, 0.80], BF = 8.42 × 1012, flute, t(84) = 6.70, p
231 < .001, d = 0.39, 95% CI [0.19, 0.59], BF = 4.81 × 106, and singing voice versions, t(84) =
232 4.74, p < .001, d = 0.28, 95% CI [0.09, 0.46], BF = 1.91 × 103. But even so, all the mean
233 ratings except one were below four, indicating the melodies were unfamiliar.
234 In sum, these results confirmed that the experimental manipulation of emotion was
235 successful, and the subjects were not familiar with the music.
237 To examine whether beauty ratings are related to brightness and harshness, a one-way
238 repeated measures ANOVA with timbre as a within-subject factor was conducted on the
240 Brightness. The data on spectral centroid met the assumption of normality, Ws ≥ 0.97,
241 ps ≥ .282, but violated the assumption of sphericity, W(2) = 0.42, χ2(2) = 32.86, p < .001.
242 Thus, we conducted the nonparametric ANOVA. The result provided strong evidence for the
243 effect of timbre, F(2, 78) = 57.31, p < .001, ωp2 = .48, 95% CI [.30, .62], BF = 1.67 × 1013.
VIOLIN IS THE MOST BEAUTIFUL 12
244 Post hoc comparisons (see Table 5) showed that spectral centroid was highest for the violin
245 version, followed by the singing voice version, and then the flute version.
246 --------------------------------------------------------
248 --------------------------------------------------------
249 Harshness. For harmonic spectral variation, the assumptions of normality, Ws ≤ 0.98,
250 ps ≤ .635, and sphericity, W(2) = 0.85, χ2(2) = 6.30, p = .043, were violated. The data
251 provided strong evidence for the effect of timbre, F(2, 78) = 86.86, p < .001, ωp2 = .46, 95%
252 CI [.28, .60], BF = 4.98 × 1017. Post hoc comparisons showed that harmonic spectral
253 variation was higher for the violin version than for the singing voice and flute versions.
254 However, there was no difference between the singing voice and flute versions. For
255 MFCCratio, the assumptions of normality, Ws ≤ 0.96, ps ≤ .135, and sphericity, W(2) = 0.51,
256 χ2(2) = 25.24, p < .001, were violated. The data also provided strong evidence for the effect
257 of timbre, F(2, 78) = 71.71, p < .001, ωp2 = .49, 95% CI [.31, .63], BF = 3.58 × 1015. Post
258 hoc comparisons showed that the MFCCratio was highest for the violin version, followed
259 by the flute version, and then the singing voice version. For attack slope, the assumptions of
260 normality, Ws ≤ 0.59, ps < .001, and sphericity, W(2) = 0.71, χ2(2) = 13.09, p = .001, were
261 violated. The data provided strong evidence against the effect of timbre, F(2, 78) = 1.18, p
263 Together, these results suggest that the violin was the brightest and the harshest timbre.
VIOLIN IS THE MOST BEAUTIFUL 13
265 Figure 1 illustrates the ratings of musical beauty for three timbre versions by music
266 and nonmusic groups. Although sphericity was met, W(2) = 0.99, χ2(2) = 0.87, p = .647,
267 normality, Ws ≥ 0.93, ps ≥ .012, and homoscedasticity, Fs(1, 83) ≥ 4.26, ps ≤ .042, were
268 violated. Therefore, the nonparametric 3 (timbre) × 2 (group) mixed ANOVA was run. The
269 ANOVA revealed strong evidence for the presence of the main effect of timbre, F(2, 166) =
270 51.83, p < .001, ωp2 = .11, 95% CI [.03, .20], BF = 1.46 × 1015, moderate evidence for the
271 absence of the effect of group, F(1, 83) = 0.42, p = .518, ωp2 = −.01, 95% CI [.00, 1.00], BF
272 = 0.11, and strong evidence for the absence of the interaction of timbre and group, F(2, 166)
273 = 0.46, p = .632, ωp2 < .01, 95% CI [.00, 1.00], BF = 0.01.
274 --------------------------------------------------------
276 --------------------------------------------------------
277 Following the main effect of timbre, post hoc analyses were conducted. There was
278 strong evidence to support that beauty ratings for the violin version were higher than for the
279 flute, t(83) = 6.15, p < .001, d = 0.52, 95% CI [0.29, 0.74], BF = 4.75 × 105, and singing
280 voice versions, t(83) = 10.10, p < .001, d = 0.85, 95% CI [0.59, 1.11], BF = 1.72 × 1013.
281 There was also strong evidence to support that beauty ratings for the flute version were
282 higher than for the singing voice version, t(83) = 3.95, p < .001, d = 0.33, 95% CI [0.12,
284 In short, these findings indicate that both music and nonmusic students rated the violin
285 as the most beautiful timbre and rated the singing voice as the least beautiful timbre.
VIOLIN IS THE MOST BEAUTIFUL 14
286 Discussion
287 In this study, to test those historical views of musical aesthetics—the human voice is
288 the most beautiful, followed by the violin, and finally the flute—we compared music and
289 nonmusic students’ ratings of perceived beauty for melodies presented in singing voice,
290 violin, and flute timbres. We found that both groups of subjects rated the violin as the most
291 beautiful, followed by the flute, and finally the singing voice.
292 The first result shows that the singing voice is the least beautiful. In other words,
293 instrumental music is more beautiful than vocal music. This result is consistent with a prior
294 study suggesting that familiar instrumental music is more beautiful than vocal music
295 (Brattico et al., 2011). However, our result goes beyond this study by showing that
296 unfamiliar instrumental music is more beautiful than vocal music without lyrics, possibly
297 due to the higher harshness of instrumental timbres. Taken together, these results suggest
298 that instrumental music is more beautiful than vocal music regardless of whether or not the
299 individual is familiar with the music and whether or not the vocal music has clear semantic
300 information, thus arguing against the idea that the human voice is the most beautiful or that
301 the human voice is more beautiful than instrumental timbre (Doni, 1640, as cited in Cypess,
302 2010; Ancelet, 1757, as cited in Dolan, 2013; Rousseau, 1765, as cited in Vélez, 2021).
303 The second result shows that the violin is the most beautiful. This result supports the
304 view that the violin is more beautiful than the flute (Doni, 1640, as cited in Cypess, 2010;
305 Ancelet, 1757, as cited in Dolan, 2013; Rousseau, 1765, as cited in Vélez, 2021). Rousseau
306 (1765) believes that the reason why the violin is more beautiful than the flute is that the
307 violin is bright and soft. Nevertheless, the present results contradict this view and suggest
308 that the reason is that the violin is brighter and harsher than the flute.
VIOLIN IS THE MOST BEAUTIFUL 15
309 Another result shows that there is no modulatory effect of musical training. This
310 result is in line with previous findings that the relationship between tonality (Nieminen et al.,
311 2011) or harmony (Jaśkiewicz et al., 2016; Müller et al., 2009, 2010) and perceived beauty
312 was not modulated by musical training. Nonetheless, our finding extends prior findings to
313 show that the relationship between timbre and perceived beauty is also not modulated by
315 There are two potential limitations of this study. One limitation relates to sample
316 representation. Although 8-year-old children have the ability to perceive beauty in music
317 (Nieminen et al., 2011), only young adult subjects (i.e., college students) were used in this
318 study, and children, adolescents, and middle-aged and older adults were not included.
319 Another limitation is related to the number of timbre types. We only used the female voice
320 and two instruments, but did not use the male voice and other instruments such as trumpet
321 from brass instruments and piano from keyboard instruments. Therefore, it is still unclear
322 whether the current results will generalize to other age groups, instruments or the singing
323 voice. Future studies need to use more age groups or timbres to increase the external
324 validity of the results. With the flourishing of neuroaesthetics (e.g., Brattico, 2019;
325 Chatterjee & Vartanian, 2014; Pearce et al., 2016; Skov & Nadal, 2022), future research on
326 experimental neuroaesthetics will reveal the nature of how timbre affects judgments of
327 beauty.
328 In conclusion, we tested the ideas that “beauty ratings for the human voice are the
329 highest, followed by the violin, and finally the flute” by asking subjects to rate the same
330 music sung or played by the human voice, violin and flute. We provide the first
331 experimental evidence that the violin is the most beautiful and the human voice the least
VIOLIN IS THE MOST BEAUTIFUL 16
332 beautiful. These findings have important practical implications: if composers and
333 performers want to produce the most beautiful music, they should give priority to the violin;
334 if audiences want to experience the most beautiful music, they should listen to the version
335 played by the violin. Timbre is the basic element of music. Future research examining the
336 role of timbre on musical beauty will enable us to have a more comprehensive
338 References
341 Anderson, S. F., & Maxwell, S. E. (2016). There’s more than one way to conduct a
344 Anglada-Tort, M., Steffens, J., & Müllensiefen, D. (2019). Names and titles matter: The
345 impact of linguistic fluency and the affect heuristic on aesthetic and value
346 judgements of music. Psychology of Aesthetics, Creativity, and the Arts, 13(3), 277-
348 Brattico, E. (2019). The neuroaesthetics of music: A research agenda coming of age. In M.
349 H. Thaut & D. A. Hodges (Eds.), The Oxford handbook of music and the brain (pp.
351 https://doi.org/10.1093/oxfordhb/9780198804123.013.15
352 Brattico, E., Alluri, V., Bogert, B., Jacobsen, T., Vartiainen, N., Nieminen, S., &
353 Tervaniemi, M. (2011). A functional MRI study of happy and sad emotions in music
355 https://doi.org/10.3389/fpsyg.2011.00308
356 Brattico, E., Brusa, A., Dietz, M. J., .Jacobsen, T., Fernandes, H. M., Gaggero, G.,
357 Toiviainen, P., Vuust, P., & Proverbio, A. M. (2020). Beauty and the brain:
358 Investigating the neural and musical attributes of beauty during a naturalistic music
360 Brydges, C. R., & Bielak, A. A. M. (2020). A Bayesian analysis of evidence in support of
361 the null hypothesis in gerontological psychology (or lack thereof). The Journals of
363 Campbell, J. I. D., & Thompson, V. A. (2012). MorePower 6.0 for ANOVA with relational
364 confidence intervals and Bayesian analysis. Behavior Research Methods, 44(4),
366 Chatterjee, A., & Vartanian, O. (2014). Neuroaesthetics. Trends in Cognitive Sciences,
368 Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Erlbaum.
369 https://doi.org/10.4324/9780203771587
370 Cohen, J. (1994). The earth is round (p < .05). American Psychologist, 49(12), 997-1003.
371 https://doi.org/10.1037/0003-066X.49.12.997
372 Cypess, R. (2010). "Esprimere la voce humana": Connections between vocal and
373 instrumental music by Italian composers of the early seventeenth century. Journal of
375 de Jong, M. A., van Mourik, K. R., & Schellekens, H. M. C. (1973). A physiological
377 https://doi.org/10.1159/000286468
378 Dolan, E. I. (2013). The orchestral revolution: Haydn and the technologies of timbre.
380 Eerola, T., Ferrer, R., & Alluri, V. (2012). Timbre and affect dimensions: Evidence from
381 affect and similarity ratings and acoustic correlates of isolated instrument sounds.
383 Eerola, T., & Vuoskoski, J. K. (2011). A comparison of the discrete and dimensional
385 https://doi.org/10.1177/0305735610362821
386 Elkin, L. A., Kay, M., Higgins, J. J., & Wobbrock, J. O. (2021). An aligned rank transform
387 procedure for multifactor contrast tests. In The 34th Annual ACM Symposium on
388 User Interface Software and Technology (pp. 754‐ 768). Association for Computing
390 Faulkenberry, T. J. (2019). Estimating evidential value from analysis of variance summaries:
393 Faulkenberry, T. J. (2021). The Pearson Bayes factor: An analytic formula for computing
394 evidential value from minimal summary statistics. Biometrical Letters, 58(1), 1-26.
395 https://doi.org/10.2478/bile-2021-0001
396 Feys, J. (2016). New nonparametric rank tests for interactions in factorial designs with
397 repeated measures. Journal of Modern Applied Statistical Methods, 15(1), 78-99.
398 https://doi.org/10.22237/jmasm/1462075500
399 Field, A. (2017). Discovering statistics using IBM SPSS statistics (5th ed.). SAGE
400 Publications.
401 Giannakopoulos, T., & Pikrakis, A. (2014). Audio features. In T. Giannakopoulos & A.
402 Pikrakis (Eds.), Introduction to audio analysis: A MATLAB approach (pp. 59-103).
404 Giannouli, V., Yordanova, J., & Kolev, V. (2022). The primacy of beauty in music, visual
405 arts and literature: Not just a replication study in the Greek language exploring the
VIOLIN IS THE MOST BEAUTIFUL 20
406 effects of verbal fluency, age and gender. Psychological Reports, 125(5), 2636-2663.
407 https://doi.org/10.1177/00332941211026836
408 Gotlieb, H., & Konečni, V. J. (1985). The effects of instrumentation, playing style, and
409 structure in the Goldberg Variations by Johann Sebastian Bach. Music Perception,
411 Hanslick, E. (1986). On the musically beautiful: A contribution towards the revision of the
412 aesthetics of music (G. Payzant, Trans.). Hackett Publishing. (Original work
414 Harms, C., & Lakens, D. (2018). Making 'null effects' informative: Statistical techniques
415 and inferential frameworks. Journal of Clinical and Translational Research, 3(2),
417 Hur, Y.-J. (2020). An empirical aesthetics of the sublime and beautiful [Doctoral
419 https://discovery.ucl.ac.uk/id/eprint/10091352/18/Hur_10091352_thesis_redacted.pd
420 f
421 Istók, E., Brattico, E., Jacobsen, T., Krohn, K., Müller, M., & Tervaniemi, M. (2009).
422 Aesthetic responses to music: A questionnaire study. Musicae Scientiae, 13(2), 183-
424 Jaśkiewicz, M., Francuz, P., Zabielska-Mendyk, E., Zapała, D., & Augustynowicz, P.
426 involving laypersons and experts. Acta Neurobiologiae Experimentalis, 76(2), 142-
428 Jeffreys, H. (1961). Theory of probability (3rd ed.). Oxford University Press.
VIOLIN IS THE MOST BEAUTIFUL 21
429 Juslin, P. N., Ingmar, E., & Danielsson, J. (2022). Aesthetic judgments of music: Reliability,
432 Juslin, P. N., & Isaksson, S. (2014). Subjective criteria for choice and aesthetic judgment of
433 music: A comparison of psychology and music students. Research Studies in Music
435 Keysers, C., Gazzola, V., & Wagenmakers, E.-J. (2020). Using Bayes factor hypothesis
438 Kim, H.-G., Moreau, N., & Sikora, T. (2006). MPEG-7 audio and beyond: Audio content
440 Lakens, D. (2017). Equivalence tests: A practical primer for t tests, correlations, and meta-
442 https://doi.org/10.1177/1948550617697177
443 Lakens, D., McLatchie, N., Isager, P. M., Scheel, A. M., & Dienes, Z. (2020). Improving
444 inferences about null effects with Bayes factors and equivalence tests. The Journals
446 Lartillot, O., Toiviainen, P., & Eerola, T. (2008). A matlab toolbox for music information
448 Data analysis, machine learning and applications (pp. 261-268). Springer.
449 https://doi.org/10.1007/978-3-540-78246-9_31
450 Lee, M. D., & Wagenmakers, E.-J. (2014). Bayesian cognitive modeling: A practical course.
452 Love, J., Selker, R., Marsman, M., Jamil, T., Dropmann, D., Verhagen, J., Ly, A., Gronau,
453 Q. F., Šmíra, M., Epskamp, S., Matzke, D., Wild, A., Knight, P., Rouder, J. N.,
454 Morey, R. D., & Wagenmakers, E.-J. (2019). JASP: Graphical statistical software
456 https://doi.org/10.18637/jss.v088.i02
457 Mair, P., & Wilcox, R. (2020). Robust statistical methods in R using the WRS2 package.
459 01246-w
460 Marin, M. M., Lampatz, A., Wandl, M., & Leder, H. (2016). Berlyne revisited: Evidence
461 for the multifaceted nature of hedonic tone in the appreciation of paintings and
463 https://doi.org/10.3389/fnhum.2016.00536
466 https://doi.org/10.3758/s13428-010-0049-5
469 McAdams, S., Douglas, C., & Vempala, N. N. (2017). Perception and modeling of affective
472 Müller, M., Höfel, L., Brattico, E., & Jacobsen, T. (2009). Electrophysiological correlates of
473 aesthetic music processing. Annals of the New York Academy of Sciences, 1169(1),
475 Müller, M., Höfel, L., Brattico, E., & Jacobsen, T. (2010). Aesthetic judgments of music in
478 Nieminen, S., Istók, E., Brattico, E., & Tervaniemi, M. (2011). The development of the
479 aesthetic experience of music: Preference, emotions, and beauty. Musicae Scientiae,
481 North, A. C., & Hargreaves, D. J. (1995). Subjective complexity, familiarity, and liking for
483 Oldfield, R. C. (1971). The assessment and analysis of handedness: The Edinburgh
485 3932(71)90067-4
486 Pearce, M. T., Zaidel, D. W., Vartanian, O., Skov, M., Leder, H., Chatterjee, A., & Nadal,
489 https://doi.org/10.1177/1745691615621274
490 Peeters, G., Giordano, B. L., Susini, P., Misdariis, N., & McAdams, S. (2011). The Timbre
491 Toolbox: Extracting audio descriptors from musical signals. The Journal of the
493 https://doi.org/10.1121/1.3642604
494 Peeters, G., McAdams, S., & Herrera, P. (2000, August 27-September 1). Instrument sound
497 Rawlings, D., & Leow, S. H. (2008). Investigating the role of psychoticism and sensation
498 seeking in predicting emotional reactions to music. Psychology of Music, 36(3), 269-
500 Rouder, J. N., Morey, R. D., Speckman, P. L., & Province, J. M. (2012). Default Bayes
501 factors for ANOVA designs. Journal of Mathematical Psychology, 56(5), 356-374.
502 https://doi.org/10.1016/j.jmp.2012.08.001
503 Rouder, J. N., Speckman, P. L., Sun, D., Morey, R. D., & Iverson, G. (2009). Bayesian t
504 tests for accepting and rejecting the null hypothesis. Psychonomic Bulletin & Review,
506 Rousseau, J.-J. (1765). Tymbre. In D. Diderot & J. l. R. d'Alembert (Eds.), Encyclopédie
507 (Vol. 16, pp. 775). Chez Samuel Faulche & Companie. http://enccre.academie-
508 sciences.fr/encyclopedie/article/v16-2453-0/
509 Rozé, J., Aramaki, M., Kronland-Martinet, R., & Ystad, S. (2017). Exploring the perceived
510 harshness of cello sounds by morphing and synthesis techniques. The Journal of the
512 https://doi.org/10.1121/1.4978522
513 Rozé, J., Aramaki, M., Kronland-Martinet, R., & Ystad, S. (2020). Cellists’ sound quality is
514 shaped by their primary postural behavior. Scientific Reports, 10(1), Article 13882.
515 https://doi.org/10.1038/s41598-020-70705-8
516 Saitis, C., & Siedenburg, K. (2020). Brightness perception for musical instrument sounds:
517 Relation to timbre dissimilarity and source-cause categories. The Journal of the
519 https://doi.org/10.1121/10.0002275
VIOLIN IS THE MOST BEAUTIFUL 25
520 Sarasso, P., Ronga, I., Pistis, A., Forte, E., Garbarini, F., Ricci, R., & Neppi-Modona, M.
524 Schiller, F. (1844). The poems and ballads of Schiller (E. B. Lytton, Trans.; Vol. 1).
526 Schubert, E., & Wolfe, J. (2006). Does timbral brightness scale with frequency and spectral
528 Seashore, C. E. (1936). The Psychology of Music. Music Educators Journal, 23(1), 24-26.
529 https://doi.org/10.2307/3384948
530 Sisti, A., Aryan, N., & Sadeghi, P. (2021). What is beauty? Aesthetic Plastic Surgery, 45(5),
532 Skov, M., & Nadal, M. (2022). The Routledge international handbook of neuroaesthetics.
534 Springer, D. G., Schlegel, A. L., & Lewis, A. J. (2021). Effects of dark and bright timbral
535 instructions on the production of pitch and timbre. Journal of Research in Music
537 Suzuki, M., Okamura, N., Kawachi, Y., Tashiro, M., Arao, H., Hoshishiba, T., Gyoba, J., &
538 Yanai, K. (2008). Discrete cortical regions associated with the musical beauty of
539 major and minor chords. Cognitive, Affective, & Behavioral Neuroscience, 8(2),
541 van Elferen, I. (2020). Timbre: Paradox, materialism, vibrational aesthetics. Bloomsbury
543 Vélez, D. V. (2021). The matter of timbre: Listening, genealogy, sound. In E. I. Dolan & A.
544 Rehding (Eds.), The Oxford handbook of timbre (pp. 22–51). Oxford University
546 Vuoskoski, J. K., & Eerola, T. (2017). The pleasure evoked by sad music is mediated by
548 https://doi.org/10.3389/fpsyg.2017.00439
549 Vuoskoski, J. K., Zickfeld, J. H., Alluri, V., Moorthigari, V., & Seibt, B. (2022). Feeling
550 moved by music: Investigating continuous ratings and acoustic correlates. PLOS
552 Wagenmakers, E.-J., Marsman, M., Jamil, T., Ly, A., Verhagen, J., Love, J., Selker, R.,
553 Gronau, Q. F., Šmíra, M., Epskamp, S., Matzke, D., Rouder, J. N., & Morey, R. D.
554 (2018). Bayesian inference for psychology. Part I: Theoretical advantages and
556 https://doi.org/10.3758/s13423-017-1343-3
557 Wilcox, R. R. (2017). Modern statistics for the social and behavioral sciences: A practical
559 Wilcox, R. R. (2022). Introduction to robust estimation and hypothesis testing (5th ed.).
561 Wilcox, R. R., Peterson, T. J., & McNitt-Gray, J. L. (2018). Data analyses when sample
562 sizes are small: Modern advances for dealing with outliers, skewed distributions,
564 https://doi.org/10.1123/jab.2017-0269
VIOLIN IS THE MOST BEAUTIFUL 27
565 Wilcox, R. R., & Serang, S. (2017). Hypothesis testing, p values, confidence intervals,
566 measures of effect size, and Bayesian methods in light of modern robust techniques.
568 https://doi.org/10.1177/0013164416667983
569 Wobbrock, J. O., Findlater, L., Gergle, D., & Higgins, J. J. (2011). The aligned rank
570 transform for nonparametric factorial analyses using only anova procedures. In
571 Proceedings of the ACM Conference on Human Factors in Computing Systems (pp.
573 https://doi.org/10.1145/1978942.1978963
575 interaction. In J. Robertson & M. Kaptein (Eds.), Modern statistical methods for
577 Yoo, H., Kang, S., & Fung, V. (2018). Personality and world music preference of
578 undergraduate non-music majors in South Korea and the United States. Psychology
580 Zhang, W., Liu, F., Zhou, L., Wang, W., Jiang, H., & Jiang, C. (2019). The effects of timbre
582 https://doi.org/10.1525/mp.2019.37.2.134
583
VIOLIN IS THE MOST BEAUTIFUL 28
584 Table 1
Age (years) 22.00 1.00 21.00 3.00 978.50 .495 .08 [−.16, .32] 0.23
Handedness 85.16 26.67 85.71 28.57 903.00 .996 < .01 [−.24, .24] 0.23
Years of education 15.00 2.00 15.00 2.00 1000.50 .379 .11 [−.14, .34] 0.24
Years of music training 10.00 8.00 0.00 0.00 1804.00 < .001 1.00 [1.00, 1.00] 4.54 × 104
586 Note. Mann-Whitney tests and its Bayesian equivalents were run with JASP.
587
VIOLIN IS THE MOST BEAUTIFUL 29
588 Table 2
591
VIOLIN IS THE MOST BEAUTIFUL 30
592 Table 3
Valence 5.37 ± 0.45 3.26 ± 0.77 5.28 ± 0.41 3.38 ± 0.71 5.08 ± 0.67 3.45 ± 0.73
Arousal 5.12 ± 0.54 3.18 ± 0.79 4.84 ± 0.58 3.19 ± 0.70 4.94 ± 0.75 3.44 ± 0.74
Familiarity 4.17 ± 1.19 3.55 ± 1.19 3.99 ± 1.22 3.38 ± 0.71 3.79 ± 1.19 3.49 ± 1.18
595
VIOLIN IS THE MOST BEAUTIFUL 31
596 Table 4
597 ANOVA Results and BF Values for the Effects of Timbre and Emotion on Three Measures
Valence
Timbre × Emotion 18.23 2, 168 < .001 .03 [.00, .09] 7.89 × 104
Arousal
Timbre 13.49 2, 168 < .001 .03 [.00, .08] 1.50 × 103
Timbre × Emotion 24.88 2, 168 < .001 .03 [.00, .09] 1.52 × 107
Familiarity
Timbre 17.21 2, 168 < .001 .01 [.00, .05] 3.42 × 104
Timbre × Emotion 11.20 2, 168 < .001 < .01 [.00, .03] 205.85
598 Note. ANOVA and BF Values are based on the aligned-and-ranked data.
599
VIOLIN IS THE MOST BEAUTIFUL 32
600 Table 5
Spectral centroid Violin vs. Singing voice 5.44 < .001 1.20 [0.57, 1.84] 5.94 × 103
Violin vs. Flute 10.71 < .001 2.37 [1.52, 3.21] 2.24 × 1010
Singing voice vs. Flute 5.26 < .001 1.17 [0.54, 1.79] 3.49 × 103
Harmonic spectral variation Violin vs. Singing voice 11.39 < .001 1.96 [1.28, 2.65] 1.30 × 1011
Violin vs. Flute 11.44 < .001 1.97 [1.28, 2.66] 1.48 × 1011
Singing voice vs. Flute 0.06 .955 0.01 [−0.41, 0.43] 0.17
MFCCratio Violin vs. Singing voice 11.53 < .001 2.33 [1.52, 3.14] 1.86 × 1011
Singing voice vs. Flute −8.56 < .001 −1.73 [−2.42, −1.05] 6.05 × 107
602
VIOLIN IS THE MOST BEAUTIFUL 33
603 Figure 1
605
606 Note. Each circle represents an individual subject; filled circles and error bars represent the