This is a archival document describing steps taken for Project 2 (Physiological and EEG Correlates of Chills) for John Chuang's Info 290 Sensors, Humans, Data, Apps class.
A. Steps so far:
In project 1, I investigated whether certain music structures tended to correspond with the experience of chills. I identified various music traits in pieces that had induced chills for me in the past, summarized briefly below.
However, in subsequent tests with myself, I found that pieces with time-dependent music structure did not reliably induce chills for me. Only one piece, Arvo Part's Spiegel im Spiegel, reliably induced chills even after repeated (dozens) listens.
1. Music Trait 1: Resolution of drawn-out tension
Samuel Barber's Adagio for Strings
The resolution occurs after a period of rising dissonance between 5:30 and 5:54, with a "gulf" of several seconds before resolution occurs. This is clearly visible from the audio spectrogram.
2. Music Trait 2: "Spectral complexity"
Jeff Goldsmith's Patch of Blue
In Patch of Blue, the combination of the high, clear whistle which brings us into the passage, and the rich timbre of strings coming in at the lower end of the spectrum around 0:20, is clearly visible from the spectrogram.
This is mirrored in Andrew Bird's Yawny at the Apocalypse, with birdsong combined with a deep, lower-frequency saturation of sound around the 0:10 mark, until a shrill pitch (rather like whale-song) is introduced at the 0:42 mark. A spectral band or gap is clearly visible stretching throughout the song after the high pitch is introduced.
3. Chosen piece: Arvo Part's Spiegel im Spiegel
Arvo Part's Spiegel im Spiegel
Arvo Part's Spiegel im Spiegel combines both traits (resolution of dissonance and spectral complexity) but in a way that much less distributed in time. From the spectrogram, we can see both dissonance and resolution, as well as the use of spectral range and complexity, occurring in repetitive phrases throughout the piece.
B. Issues faced:
Issue 1: Music with time-dependent music structure did not reliably induce chills
Interestingly, the pieces with a more pronounced time-dependent music structure did not reliably elicit chills. This coheres with research that suggests unexpected changes in music are more likely to induce chills. Because the ability of Bird's Yawny or Goldmsith's Patch of Blue to induce chills depended on specific points in the piece, and came to be expected, this might explain why Spiegel was the only piece that consistently induced chills. Its regularity, in some ways, prevented pre-empting of a specific section of the music, since the music traits occurred continually throughout the piece rather than only at one specific juncture.
Issue 2: Self-reporting of chills
Another issue faced with EEGs was that having participants record chills were likely to interfere with EEG readings. In order to find a way to reliably detect chills from other physiological data, and not just self-reporting, I investigated the literature to find other physiological data that were correlated with the experience of chills. Two physiological phenomena - piloerection (goosebumps) and elevated GSRs) were most promising. In tests done on myself with a smartphone camera, I did not find visible differences in piloerection during the experience of chills, on my upper and forearms, neck and thigh. I decided to investigate more closely whether changes in GSRs could reliably indicate the experience of chills.
C. Research Plan:
Going forward, I had two goals, the second of which was a back-up option in case changes in GSRs could not reliably pick up the experience of chills.Original Goal: If chills can be reliably identified by GSR, use GSR to divide EEG data into chills and no-chills, and perform FFT to compare frequencies.
Alternative Goal: Examine physiological data in pieces with time-dependent music structure.
D: Methodology:
To assess whether GSR readings could detect chills reliably, I first wrote a simple recorder using Pygame's key logging capabilities, where the participant pushes the space bar when he experiences the onset of a "chill" and releases once it ends.
I then, as before, listened to musical pieces in a dark room with eyes closed, with GSR and ECG sensors from the Bitalino attached to the right palm and chest. The goal was that if GSR readings were found to be highly correlated with self-reported chills, I can use GSRs to categorize EEG readings without self-reporting.
E. Exploratory Analysis:
My initial analysis was very intriguing. In a first run, the following results were obtained, with the line in blue being GSR levels, and the green and red lines signalling the beginning and end of a chill respectively.
As can be seen, for the first 13 or so chills in the piece, all but one chill could be tied to a sudden spike in GSR readings. The last 5 chills however, did not correspond to any upsurge in GSR readings.
Excited by these results, I conducted several more tests subsequently, but found the effect had disappeared. In these later tests, which occurred immediately after the first, the occurrence of chills no longer occurred with notable spikes in GSR activity. Below is an example of these later readings. 
These exploratory tests indicated that GSRs could not reliably detect chills, though they offer some interesting phenomenon and hypotheses for further testing. One possible hypothesis is whether GSRs tends to more highly correlated and in sync with the experience of chills in a lower state of arousal, as the chills ceased to match up with spikes in GSR levels close to the end of the first trial and not at all in subsequent trials.
Given that we cannot reliably detect chills with GSRs, I explored the alternative goal of examining physiological data with a piece with a more time-dependent music structure - in this case, Samuel Barber's Adagio for Strings.
Spectrogram for Samuel Barber's Adagio for Strings:
In the diagrams below, the green line marks the period of rising dissonance around 270 seconds and ending with the resolution at 410s.
ECG readings:
EDA readings:
ECG readings do not reveal unusual activity in the period of rising dissonance and resolution. EDA readings, however, do show a rise in EDA levels in the later half of the period of rising dissonance, before falling in the final section of the music occurring after the resolution around the red line. There is no obvious spikes or dips around either the "gulf" before resolution occurs (in anticipation), or during the resolution itself. One interesting direction would be to compare results for participants who had not heard the piece before or are less familiar with the piece, to see whether the pivotal resolution of the piece might evoke a more pronounced response.
G. Next Steps:
Despite failing to obtain a reliable indicator of chills from physiological data, these results present some interesting directions for next steps.
Many of the relationships observed would be interesting as applied to a larger sample, without extending to EEG readings. For example, simply having participants record their experience of chills (using the keylogger) together with GSR readings would provide a stronger basis for verifying the "saturation" hypothesis, i.e. whether GSR readings tend to match up with experienced chills in a lower state of arousal.
Similarly, the "unexpectedness" of music traits could also be examined for a larger sample, by whether they have heard the piece previously. One further interesting space for analysis is by altering various musical traits of the same piece of music (etc. by increasing spectral complexity by the addition of sound effects or shifting a voice one key up), and seeing whether an increase in chills tends to be associated with altering the music in some fashion. For the musician/composer, this would provide better data as to how chills can be evoked at key points in the music, which may apply especially to soundtracks for film and television.
