See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/221494070

Understanding Emotion in Raag: An Empirical Study of Listener Responses

Conference Paper in Lecture Notes in Computer Science · August 2007

DOI: 10.1007/978-3-540-85035-9_7 · Source: DBLP

CITATIONS READS

52 2,997

2 authors, including:

Parag Chordia

28 PUBLICATIONS 870 CITATIONS

SEE PROFILE

All content following this page was uploaded by Parag Chordia on 20 March 2014.

The user has requested enhancement of the downloaded file.

 UNDERSTANDING EMOTION IN RAAG: AN EMPIRICAL STUDY OF
LISTENER RESPONSES

Parag Chordia and Alex Rae

Georgia Institute of Technology, Department of Music
840 Mc Millan St., Atlanta GA 30332
{ppc,arae3}@gatech.edu

ABSTRACT theoretically thousands of scale types; in practice, how-

ever, raags conform to a much smaller set of scales, and
A survey of emotion in North Indian classical music was
many of the most common raags share the same set of
undertaken to determine the type and consistency of emo-
notes.
tional responses to raag. Participants listened to five one-
By building phrases as described above, a tonal hier-
minute raag excerpts and recorded their emotional responses
archy is created. Some tones appear more often in the
after each. They were asked to describe the emotions each
basic phrases, or are sustained longer. Indian music the-
excerpt evoked and then to adjust six different sliders indi-
ory has a rich vocabulary for describing the function of
cating the degree to which they felt the following: happy,
notes in this framework. The most stressed note is called
sad, peaceful, tense, romantic, longing. A total of 280 re-
the vadi and the second most stressed, traditionally a fifth
sponses were received. We find that both free-response
or fourth away, is called the samvadi. There are also less
and quantitative judgments of emotions are significantly
commonly used terms for tones on which phrases begin
different for each raag and quite consistent across listen-
and end. A typical summary of a raag includes its scale
ers. We hypothesized that the primary predictors of emo-
type (that), vadi and samvadi. A pitch-class distribution
tion in these excerpts would be pitch-class distribution,
(PCD), which gives the relative frequency of each scale
pitch-class dyad entropy, overall sensory dissonance, and
degree, neatly summarizes this information.
note density. Multiple regression analysis was used to
The performance context of raag music is essentially
determine the most important factors, their relative im-
monophonic, although vocalists will usually be shadowed
portance, and their total predictive value (R2 ). The fea-
by an accompanying melody instrument. The rhythmic
tures in combination explained between 11% (peaceful)
accompaniment of the tabla is also present in metered sec-
and 33% (happy) of response variance. For all models,
tions. There is usually an accompanying drone that sounds
a subset of the features were significant, with the inter-
the tonic and fifth using a harmonically rich timbre.
play between “minor” and “major” scale degrees playing
an important role. Although the explanatory power of the
current models is limited, the results thus far strongly sug- 2. RELATED WORK
gest that raags do consistently elicit specific emotions that
are linked to musical properties. The responses did not The emotional qualities of raags have been discussed ex-
differ significantly for enculturated and non-enculturated tensively in NICM music theory. Traditionally, this mu-
listeners, suggesting that musical rather than cultural fac- sic is said to evoke seven basic emotions: sadness, ro-
tors are dominant. mance, peace, strength/courage, anger, dispassion, devo-
tion (Bhatkande 1934). In theory, each raag elicits a unique
1. BACKGROUND emotional state (rasa) consisting of one or more of these
basic emotions.
Raag literally means “that which colors the mind”. It For Western music, a variety of studies have been un-
is a melodic abstraction around which almost all North dertaken since the late 19th century to define what emo-
Indian classical music (NICM) is organized. A raag is tions music elicits, and to understand the musical factors
most easily explained as a collection of melodic gestures that underly them. These studies have followed a basic
and a technique for developing them. The gestures are experimental paradigm in which musical excerpts are pre-
sequences of notes that are often inflected with various sented to listeners who are then asked to respond by ver-
micro-pitch alterations and articulated with an expressive bally describing their emotional states, rating emotions on
sense of timing. Longer phrases are built by joining these a quantitative scale, or some other measurement. Musi-
melodic atoms together. cal stimuli are chosen so that certain characteristics can
NICM uses approximately one hundred raags, of which be systematically varied, such as timbre, tempo and tonal-
fifty are common. Despite micro-tonal variation, the tones ity. Correlations between responses and musical structure
in any given raag conform to a subset of the twelve chro- are then observed, and are interpreted to account for some
matic pitches of a standard just-intoned scale. There are aspect of the listener reactions.
 A common pitfall in such experiments is the fundamen- C D[ D E[ E F F] G A[ A B[ B
tal difficulty of isolating the factor that is being tested. Yaman • • • • • • •
Real musical excerpts rarely vary in just one parameter, Desh • • • • • • •
while artificial stimuli that can be systematically varied Khamaj • • • • • • • •
often will not generalize, lacking the “ecological validity” Darbari • • • • • • •
of real music. Most importantly, the difficulty of quanti- Marwa • • • • • •
fying a person’s actual emotional state, either through ob-
servation or by eliciting verbal or other responses, makes Table 1. Summary of scale degrees used by raags in
such research extremely challenging. Despite this, the database. Notes are listed with C as the tonic.
field has produced a body of work that suggests consistent
relationships between musical structure and emotional re-
sponse (Balkwill and Thompson 1999). In general, two and the raag excerpts can be heard at http://www.
broad categories of emotional responses have been found, paragchordia.com/research. Some basic demo-
relating to valence (happy-sad) and activity (vigorous-calm). graphic information was collected, specifically age, sex,
For example, fast tempo has been shown to correlate con- level of musical training (if any), and familiarity with NICM
sistently with positive valence and activity; the same has (if any). For each excerpt, there was both a free response
been shown for major (versus minor) tonality. A thorough with the directions to “describe how it makes you feel in
overview of the field of emotion and music can be found as much detail as possible,” and a series of six value slid-
in Juslin and Sloboda (2001). ers (ranging from zero to one hundred) corresponding to
Although the connection between raag and emotion is pre-chosen emotional adjectives: happy, sad, tense, peace-
assumed to be systematic by nearly all enculturated lis- ful, romantic, and longing. The adjectives were chosen
teners and performers, very little empirical work relating based on preliminary research by the author suggesting
to emotion in NICM has been published. In one study, that these might best capture the dominant emotional axes.
NICM musicians were asked to play short excerpts that Further, the first four roughly represent the valence and
conveyed either joy, sadness, anger, or peace. Listen- activity judgments discussed above.
ers not familiar with NICM were asked to identify the Participants were asked to “adjust the slider to indicate
predominant emotion. The recognition of intended emo- how well you feel that the word applies to the raag ex-
tions was mostly successful except for peace (Balkwill cerpt you just heard.” The order in which each of these
and Thompson 1999). sliders appeared was randomized for each raag presented.
Each participant was presented with the same set of five
raags in a random order. The specific excerpt for each
3. MOTIVATION raag was chosen at random from three possibilities, one
played on sarod and two on sitar, both traditional plucked
The current work seeks to characterize the emotional re-
string instruments. All excerpts were approximately one
sponses of listeners to different raags and the extent to
minute long, unaccompanied, and from alap, a slower,
which listeners report similar experiences; furthermore, if
arrhythmic introductory section of a raag performance.
responses differ systematically by raag, it seeks to iden-
Participants had the option of replaying each segment an
tify some of the musical factors that underly these re-
unlimited number of times. Without knowing in advance
sponses. In NICM, all pitch-related activity is structured
how much data we would be able to collect, it was impos-
by the choice of raag. Each raag is traditionally asso-
sible to determine an appropriate dimensionality for the
ciated with several basic emotions. For example, Raag
data we sought. Therefore, an attempt was made to strike
Yaman is said to evoke peace, happiness, and devotion.
a balance, leading to the decision to limit the number of
Performers and listeners of NICM often report that the
slider-values to six and include a free-response section as
essence of raag-based music is the evocation of mood
well. This was also the reasoning that determined on the
brought about by one or more related emotions, and that
one hand the presentation of the same five raags to each
different raags reliably elicit different emotional states. In
participant, while generalizing slightly by the inclusion of
many cases, however, the more traditional descriptions are
three audio files for each raag.
not accepted, or are thought of as rudimentary. Despite the
central role of emotion in raag theory, almost no empiri-
cal work has been undertaken to systematically analyze
listener response to different raags. 5. SURVEY RESULTS

4. SURVEY DESIGN A total of 280 survey responses were collected. Partici-

pants were recruited from online NICM forums, as well as
For the purpose of collecting a large number of responses, through the Georgia Institute of Technology music depart-
we created an online survey in which participants were ment. 29% of the respondents were female and 71% male,
asked to respond to five raag excerpts in terms of “how with a median age of 24. The respondents described their
listening to each raag made you feel.” The survey can be familiarity with NICM as “none” (11%), “a little”(24%),
found at http://www.paragchordia.com/survey “somewhat”(29%), “very”(32%) and “expert”(4%).
Figure 1. Summary of quantitative survey responses. Box and whisker plots for each emotion are shown. The 25th to
75th percentile is indicated by the box, the median by the horizontal line in the middle of the box. The range of the data
is shown by the whiskers with outliers notated with hatch marks. It can be seen that Desh and Khamaj form one reponse
cluster, and Marwa and Darbari another.

5.1. Content Analysis of Free Responses man was equally associated with “melancholy & sadness”
and “joy & happiness.” The results show that, although
The descriptive language of the free responses indicated each raag has a strong valence, a variety of other emo-
the intensity and complexity of listeners’ emotions. The tions are also present, suggesting that listeners’ responses
style ranged from lists of words and phrases describing are indeed complex.
simple emotions to lengthy and even poetic passages. A
sampling is shown in Table 4.
5.2. Quantitative Responses
Due to the fact that many of the free responses con-
sisted of fragments or lists of words, highly sophisticated Figure 1 shows the distribution of quantitative (slider) re-
content analysis was not particularly relevant; instead, we sponses for each emotion and raag. For happy, sad, tense,
focus on lower-level analysis, primarily word histograms. and romantic, three raag clusters are apparent. Desh and
Initially, simple histograms of terms appearing in each Khamaj are strongly positively valenced (happy, roman-
free response section were tabulated. Limiting the results tic and not tense) and Marwa and Darbari are negatively
to relevant descriptive terms, the most common for each valenced (sad, not romantic, tense) with Yaman falling be-
raag were collected and are shown in Table 2. Standard tween these poles. Multiple comparison of means con-
variations of each word (e.g. “happy” and “happiness”) firms that differences between these three clusters for these
were treated as examples of the same root word. three emotions are significant (p < .05) but that differ-
Terms were also grouped according to various cate- ences within clusters are not significant. The “longing”
gories of meaning, in an attempt to both give a broader values show no statistically significant differences by raag,
view of the responses and to capture information buried however all the raags clearly brought out this emotion.
in the “long tail” of infrequently used but clearly related For the“peaceful”emotion there are again three clusters:
terms. For example, words such as “bliss”, “exultation”, Marwa is least peaceful followed by Darbari and then a
“happy”, “joyful”, and “ecstasy” were grouped together cluster containing Yaman, Khamaj and Desh.
under “Joy & Happiness”. Relevant semantic groupings
were based on a modified version of a standard content 6. FEATURE SELECTION
analysis dictionary. As before, the most common occur-
rences were tabulated and the results are displayed in Ta- In order to explore possible predictors for these emotional
ble 3. reactions, several features were extracted from the audio
The tables clearly show a common pattern of responses excerpts. Some of these features have previously been
for certain raags. Raags Khamaj and Desh are associated shown to be highly effective in raag classification (Chor-
with positively valenced concepts such as “joy & happi- dia and Rae 2007), thus we sought to investigate their abil-
ness,” whereas Darbari and Marwa were consistently as- ity to explain emotional responses. We hypothesized that
sociated with “melancholy & sadness.” Interestingly, Ya- survey responses would in part be predicted by the pitches
 Darbari Desh Khamaj Marwa Yaman
Freq. Term Freq. Term Freq. Term Freq. Term Freq. Term
75 sad 59 happy 53 happy 74 sad 46 sad
19 peaceful 41 peaceful 36 peaceful 16 serious 42 peaceful
14 deep 19 romantic 31 sad 14 good 38 happy
13 calm 17 sad 24 romantic 14 peaceful 13 good
10 longing 15 soothing 13 soothing 14 longing 12 calm
10 serious 11 nice 10 good 13 deep 11 longing
10 beautiful 10 pleasant 9 longing 10 tense 8 serious
9 good 10 patriotic 9 love 7 emotional 7 romantic
8 happy 9 playful 9 relax 7 happy 7 relax
5 sombre 9 remind 6 calm 6 soothe 6 devotional

Table 2. Simple word histograms for each Raag

Darbari Desh Khamaj Marwa Yaman

Frq. Semantic Cat. Frq. Semantic Cat. Frq. Semantic Cat. Frq. Semantic Cat. Frq. Semantic Cat.
134 melancholy & sadness 115 joy & happiness 103 joy & happiness 133 melancholy & sadness 71 melancholy & sadness
21 joy & happiness 41 love & affection 54 melancholy & sadness 28 tension 69 joy & happiness
16 pains 32 melancholy & sadness 37 love & affection 15 sympathy & compassion 16 emotions
15 emotion 14 pleasures 13 emotions 12 pains 16 love & affection
12 beauty 13 composure 10 composure 12 joy & happiness 12 beauty
11 tension 12 beauty 9 pleasures 11 emotion 9 composure
11 love & affection 11 hope & optimism 7 passions 10 love & affection 8 tension
10 sympathy & compassion 6 emotion 7 hope & optimism 7 anger & indignation 8 pleasures
8 despair & resignation 6 desires 5 desires 7 emotionalities 6 passions

Table 3. Semantic category histograms for each Raag. Words with a similar meaning were assigned to a common
semantic category.

“Emptiness. I’m tumbling down a deep abyss. Weightless, then maybe. Dark and surrounding. Fall.”
Darbari:
“It feels like pain, an agony that is long lasting and is happening at the time that the music is playing, a time
of hardship.”

“Absolutely fresh.... Clearing all the bad thoughts... Gives fresh meaning to life. Sounds serene. Very
Desh:
Peaceful i felt.”
“Pure, unblemished...Something white as milk, offering to take you in, clean all your sins...Compassionate
and loving, but in a distant sort of way. ”

“It so reminds me of the blossoming of flowers and prosperity... the happy chuckles of newborns... the
Khamaj:
pride of their mothers... ”
“spring time. birds chirping. sun is shining. love. a mother is telling her child a story. the child is smiling.
good times are upon us.”

“my reaction to this raaga was almost sexual. I felt desire. I felt the urge to look beautiful and dress up in
Marwa:
heavy gold jewelery. felt very aware of my body. at times I felt a strange anger and the desire to control
somebody else. i felt very powerful. i felt like a woman.”
“very depressing...felt like crying...someone was going away... describes life in some way....the ups and
downs. ”

“I feel like a butterfly. The wind is streaking past me, and colors are awash in the air. Melodious colors.”
Yaman:
“A combination of moods. It looks relaxed most of the time and ruminating about something. It seems to
turn a little angry occasionally as though an unhappy event was inadvertently recollected.”

Table 4. A sample of free responses

used and their relative weights, as represented by PCDs. for each note. The label of the corresponding chromatic
In particular, it was expected that the relative prevalence pitch was assigned to that note. This process dealt quite
of certain major and minor intervals would correspond to effectively with variations due to micro-pitch structure, at-
positively and negatively valenced responses. tacks, and errors by the detection algorithm. The octaves
Further, it was thought that sequential pairs of pitches, were folded into one as with the PCDs. The pitch-classes
pitch class dyad distributions (PCDDs), would have a no- were then arranged in groups of two (bi-grams), or in mu-
ticeable correlation. Extrapolating from Huron (Huron sical terms, dyads. The entropy of the PCDD for each
2006), we hypothesized that the degree of flexibility in excerpt was then calculated.
PCDDs would correspond to a sense of tension and long-
ing. Flexibility was calculated as the entropy of the PCDD
6.4. Sensory Dissonance
distributions, where a low entropy would correspond to an
excerpt having highly determined, or leading, note transi- The presence in a sound of partials which fall within the
tions, possibly leading to a greater sense of longing. critical band for frequency resolution is the key indicator
A number of other features were considered as well, of dissonance, according to the tonotopic theory of sen-
notably the overall dissonance of a given excerpt, calcu- sory dissonance. A value for this was calculated by per-
lated as the sensory dissonance. This feature was thought forming pairwise comparisons between all detected par-
likely to correspond to negative emotions, although as cal- tials in each segment, and weighting the detected sub-
culated it might give excessively large values to spectrally critical-band intervals by their relative amplitudes. This
rich sounds. The overall note density and spectral cen- calculation was performed using an algorithm developed
troids for each excerpt were also determined. by Kameoka and Kuriyagawa (Kameoka and Kuriyagawa
1969).
6.1. Pitch Detection and Pitch Class Distributions
Pitch detection was done using a version of the Harmonic 6.5. Timbral Features
Product Spectrum (HPS) algorithm (Sun 2000). Each seg- The center of mass in the frequency domain (spectral cen-
ment was divided into 40 ms frames, using a Gaussian troid) was calculated for each excerpt.
window. The frames were overlapped by 75%, so that the
pitch was estimated every 10 ms.
The PCDs were calculated by taking histograms of the 7. ANALYSIS
pitch-tracks. The bins corresponded to each note of five
octaves of a chromatic scale centered around the tonic for An initial demographic analysis showed that responses
that segment. Specifically, the ratios of the just-intoned were not significantly different by age, sex, or familiarity
scale and the tonic frequency were used to calculate the with Indian music.
center of each bin, and the edges were determined as the Multiple regression analysis was undertaken to deter-
log mean. The five octaves were then folded into one and mine how well the PCDs, subjective dissonance, and sen-
the values normalized to create a pitch-class distribution. sory dissonance features explained the emotion ratings. A
This nulled any significance of octave errors in the HPS linear model was built using these features for each of the
algorithm. quantitative emotional response types (with the exception
of “longing”, which was excluded because no statistically
6.2. Note Density significant differences were found between raags). For
example, the slider values for “happy” were modeled as
The average note density for each excerpt was calculated
from detected note onsets. Onsets were found by thresh- happyi = β1 xi1 + β2 xi2 + β3 xi3 + + βn xin (1)
olding a complex detection function (DF) (Duxbury, Bello,
Davies, and Sandler 2003). The excerpt was divided into where xik is the kth feature and βki is the corresponding
128 sample regions, overlapped 50% using a rectangular regression coefficient for the ith observation. Stepwise re-
window, and the DFT of each region was computed and gression was performed to determine the subset of fea-
used to construct the DF. Adaptive thresholding, propor- tures to be included in the final model.
tional to the median over a sliding window, was used to Tables 5 and 6 show the beta values, that is the stan-
choose the peaks to be labeled as onsets. dardized regression coefficients, for the each of the fea-
tures, as well as the R2 statistic for the total model. All
6.3. Pitch Class Dyad Distributions coefficients presented have p values of less than .01. All
variables in the model were normalized to have a mean of
Pitch class dyad distributions were calculated for each ex- zero and a standard deviation of one, thus avoiding diffi-
cerpt. The detected onsets were used to segment the pitch- culties in comparing the scales of the independent vari-
tracks into notes. Each note was then assigned a pitch- ables. The R2 statistic measures the total variance ac-
class label: first the raw pitch estimates were discretized counted for by the model and varies between 0 (random)
by assigning to each the center value of the bins defined and 1 (perfect fit). Of course, if measurement error is high,
for the pitch histogram, and then the mode was calculated then the model will not be meaningful. The individual
 Figure 2. Pitch-class distributions for raags Desh and Khamaj

correlation between each feature and the emotion, which features that give the greatest incremental contribution to
varies between -1 and 1, is also shown. explaining the variance of the dependent variable. In this
way, only one of a set of highly correlated features is typ-
Additionally the variance inflation factor (VIF) for each
ically included. Nevertheless, highly correlated variables
feature is shown, giving a measure of the multicollinear-
may remain in the model, making VIF useful for interpre-
ity of the independent variables. This occurs when one of
tation.
the variables is well approximated by a linear combination
of the other variables, as is likely to be the case for PCD Table 5 shows that “happy” and “sad” are modeled best,
values. The VIF is calculated by regressing each variable with R2 values of .34 and .28 respectively, whereas the
using the remaining independent variables, and is com- model is only weakly predictive in the case of “peace-
1 2
puted as 1−R 2 . If the R value is high, it indicates that the ful” (.11) and “tense” (.16). Unsurprisingly, “happy” re-
variable is multicollinear. In typical applications, VIF val- sponses are negative correlated with the minor 2nd (β =
ues above a threshold of 10 indicate that a variable should −.13), minor 3rd (β = −.16), minor 6th (β = −.43), and
be removed. Interpretation of models with multicollinear minor 7th (β = −.37). Surprisingly, the Major 3rd is also
variables is difficult as the relative contribution of each negatively correlated with a β value of -.13. As mentioned
cannot easily be read from the beta values and the reliabil- before, however, the Major 3rd has a high VIF value; ex-
ity of the estimate decreases (the size of the confidence in- amination of the correlation coefficients shows that it is
terval increases for that coefficient). In some cases, multi- strongly correlated with the Perfect 4th (.84), which is pos-
collinearity can reverse the expected sign of the regression itively related to “happy” responses. “Sad” responses show
coefficient. For example, in Table 5 in the “happy” model, an opposite pattern, positively related to the minor 2nd (β =
the Major 3rd coefficient is negative despite a strong posi- .14), minor 3rd (β = .35), and minor 6th (β = .20),
tive correlation; it can be seen, however, that the variable and negatively to the Perfect 4th (β = −.20). Many of
is highly multicollinear, with a VIF of 7.74. Because scale the other models loosely conform to the idea that raags
tones in raags may be highly correlated, it is important to with “minor” intervals are negatively valenced (sad, tense)
be aware of this potential confound. Stepwise regression and raags with “major” intervals are positively valenced
partially avoids this problem by automatically selecting (happy, peaceful).
 “Happy” Regression Model — Total R2 = .34 “Sad” Regression Model — Total R2 = .28
feature β corr VIF feature β corr VIF
minor 2nd -0.13 -0.29 4.23 minor 2nd 0.14 0.26 2.11
minor 3rd -0.16 -0.09 1.86 Major 2nd -0.16 -0.14 1.91
Major 3rd -0.13 0.38 7.74 minor 3rd 0.35 0.09 2.12
Perfect 4th 0.13 0.37 5.66 Perfect 4th -0.20 -0.29 1.66
minor 6th -0.43 -0.25 6.02 minor 6th 0.20 0.28 1.36
Major 6th -0.35 -0.28 2.33 Major 6th 0.20 0.25 2.29
Major 7th -0.37 -0.03 8.11 spectral centroid -0.18 -0.10 1.98
sensory dissonance 0.13 0.06 2.74 note density -0.15 -0.11 1.65
PCDD entropy 0.18 0.38 2.27

Table 5. Summary of regression models for “happy” and “sad”. For each feature we report the standardized regression
coefficients, the bivariate correlation, and the variance inflation factor measure of multicollinearity.

“Peaceful” Regression Model — Total R2 = .11 “Tense” Regression Model — Total R2 = .16
scale deg beta corr VIF scale deg beta corr VIF
Major 2nd 0.22 0.11 3.32 minor 2nd 0.10 0.25 1.92
minor 3nd -0.24 -0.05 2.76 Major 2nd -0.21 -0.15 1.71
Major 3rd 0.22 0.22 4.69 minor 3rd 0.24 0.08 1.55
Perfect 4th 0.15 0.23 4.63 Major 3rd -0.14 -0.26 1.45
Perfect 5th 0.15 0.10 1.20 Perfect 5th -0.20 -0.18 1.74
Major 7th 0.23 0.03 1.88 minor 6th 0.14 0.12 1.45
PCDD entropy -0.16 0.19 3.46

Table 6. Summary of regression models for “peaceful” and “tense”

The entropy of the PCDD distribution was a signifi- Full PCD only Minor
cant factor in the “happy” and “peaceful” models, posi- Happy 0.34 0.27 0.07
tively related in the former and negatively related in the Sad 0.28 0.22 0.06
latter. We had hypothesized that low PCDD entropy val-
ues would correspond to tension and longing and higher
Peaceful 0.10 0.10 0.02
values to a greater sense of flexibility. The effect shown Tense 0.16 0.13 0.05
here is weakly consistent with this prediction. Romantic 0.21 0.18 0.08
It is important to note that the relationship between
Table 7. Comparison of three regression models. The
PCDD entropy and the emotional characteristics we are
total adjusted R2 value for each model is shown.
testing is likely non-linear. Values outside the range rep-
resented in this study might well be expected to elicit dif-
ferent reactions. One might expect very low values, corre- which allows comparison between models with a different
sponding to predictability and repetition, to elicit “peace- number of independent variables, is reported in Table 7.
ful” and “sad” reactions. Very high values, on the other The second model considered only the PCD features. The
hand, might correspond to unpredictability and hence elicit full model was significantly more explanatory. Although
feelings of “tension” and “stress”. However, in the mid- the full model suggested that a feature that combined the
dle of the range, where most of the musical excerpts used minor 2nd , minor 3rd and minor 6th in total measure of
here lie, low entropy corresponds to raags with a relatively “minorness” might capture most of the information in the
fixed phraseology, creating a sense of tendency rather than PCD, this was not the case. The PCD features explained
repetition, and high entropy corresponds to raags with a an additional 10-20% of variation as compared with the
greater degree of flexibility, conveying more variability single “minor” feature.
than instability. If correlations observed in the above mod-
els are valid, this is the most likely explanation.
8. DISCUSSION
Two additional models were developed to see if a more
parsimonious explanation of the data could be given. The Survey responses have shown that different raags evoke
first used the total strength of the minor 2nd , minor 3rd , a clearly differentiated set of emotional reactions. Free
and minor 6th in the raag as the independent variable, a responses tended to cluster strongly in particular adjecti-
measure of the total degree of “minorness”. The adjusted val categories based on raag, and quantitative responses
R2 , were significantly different by raag for all emotions ex-
n−1 cept “longing”. Thus, a substantial step has been taken
Ra2 = 1 − (1 − R2 ) , (2) towards empirical verification of the nature and reliabil-
n−p−1
 ity of emotional responses to raag. Importantly, responses approach to onset detection for audio segmenta-
did not vary systematically by familiarity with NICM sug- tion. In Proc. of the 4th European Workshop on
gesting that listeners were not simply referring to cultur- Image Analysis for Multimedia Interactive Services
ally determined concepts, but responding to underlying (WIAMIS–03), London, pp. 275–280.
features of the music. Huron, D. (2006). Sweet Anticipation: Music and the
The analysis, although preliminary, suggests that re- Psychology of Expectation. MIT Press.
sponses are in part attributable to pitch-class statistics; the
prevalence of certain scale degrees is useful in predicting Juslin, P. N. and J. A. Sloboda (2001). Music and
the valence of the emotional responses. The data suggest Emotion: Theory and Research. Oxford University
that the entropy of the PCDD and the spectral centroid are Press.
also important. These are undoubtedly just a few of the Kameoka, A. and M. Kuriyagawa (1969). Consonance
many factors that influence listener responses. As more theory, part i: Consonance of dyads. Journal of the
data are collected it will be possible to more fully exam- Acoustical Society of America 45(6), 1451–1459.
ine other factors. Sun, X. (2000). A pitch determination algorithm based
It is important to note that these models are currently on subharmonic-to-harmonic ratio. In In Proc. of
merely suggestive. In none of the cases were they highly International Conference of Speech and Language
predictive, with a maximum of 34% of the variance ac- Processing.
counted for. Because the goal here was explanatory rather
than classificatory, the models were not verified on an
independent data set. As with any task that forces re-
spondents to verbalize primarily non-verbal mental states,
there is significant measurement error due to the inherent
unnaturalness of the task and an imperfect ability to map
the verbal space. It is also possible that much of the true
emotional feel of the music is lost in the projection onto
simple emotions such as “happy” and “sad”. Although it
is likely that some aspect of raags can be effectively cap-
tured by mapping onto these axes, it is also likely that it is
a gross simplification of the actual emotional experience.

9. CONCLUSIONS AND FUTURE WORK

We have reported the results of the first empirical survey

of listeners’ emotional reactions to raag music. We have
established that responses exhibit clear patterns and have
identified several musical features which partially explain
them. Future work will consider a larger set of raags
and provide listeners with more dimensions along which
to evaluate their emotional experiences. As more data is
collected, it will be possible to test a greater number of
musical features and more robustly assess their validity.

References
Balkwill, L. L. and W. F. Thompson (1999). A cross-
cultural investigation of the perception of emotion
in music: Psychophysical and cultural cues. Music
Perception 17, 43–64.
Bhatkande, V. (1934). Hindusthani Sangeet Paddhati.
Sangeet Karyalaya.
Chordia, P. and A. Rae (2007). Raag recognition us-
ing pitch-class and pitch-class dyad distributions.
In Proceedings of International Conference on Mu-
sic Information Retrieval.
Duxbury, C., J. P. Bello, M. Davies, and M. Sandler
(2003). A combined phase and amplitude based

View publication stats