After years of unexplicable failure, I’ve finally gotten numpy and scipy to play nice on my computer. (Anaconda finally installed properly once I moved to Yosemite; who knows what was breaking the system before…) So now I can finally start converting my matlab toolbox for Activity Analysis to open source python, and in the process, and share analyses online with ipython notebook. Whose excited? well I am. I think it will make it easier to follow what the calculations and inferences, particularly to those who aren’t inclined to download a toolbox and run a demo on their own machines.
There are two comically extreme positions on how music (or really any stimulus) affects observers. At one end, the position that all of our experiences are equivalent, dictated by the common signal, at the other, individual subjectivities make our impressions and reactions irreconcilable. In studying how people respond to music, it’s obvious that the reality lies somewhere in the middle: parts of our experience can match that of others, though differences and conflicts persist. I’ve spent years developing this thing called activity analysis to explore and grade the distance between absolute agreement and complete disarray in the responses measured across people sharing a common experience.
As people attend to a time varying stimulus (like music) their experience develops moment by moment, changes prompted by events in the action observed. What we have, in activity analysis, is a means of exploring and statistically assessing how strongly the shared music coordinates these changes in response. So if we are tracking smiles in an audience during a concert, we can evaluate the probability that those smiles are prompted by specific moments in the performance, and from there have some expectation of how another audience may respond.
If everyone agreed with each other, this would not be necessary, and if nothing was common between listeners’ experience, this would not be possible. Instead empirical data appears to wander in between, and with that variation comes the opportunity to study factors nudging inter-response agreement one way or the other. We’ve seen extreme coherence, that of the crowd singing together at the top of their lungs in a stadium saturated with amplified sound, and polite but disoriented disengagement is a common response to someone else’s favourite music. We need to test the many theories on why so many different response (and distributions of responses) arise from shared experiences, and Activity Analysis can help with that. Finally.
Here is hoping I can get back to sharing examples of what this approach to collections of continuous responses makes possible. The data and analyses have been waited too long already.
I’ve got a new paper out, with Mary Farbood (first author) about ratings of musical tension to an interesting example of romantic lieder, Schubert’s Morgengruss. The link is not to the performance we worked with, which was the Pears and Britten recording, but I like this interpretation too.
My contribution is in the comparison between verses, identifying significant moments of tension rating increases and decreases which differed between verses, and discussing how that might be related to the singer’s articulation, contrast between successive verses, and other factors often overlooked in continuous parametrizations of musical stimuli. While displaying some of what activity analysis can do, numerically, it was also fun to put on my music theory hat to interpret what might be influencing listeners continuous ratings of tension.