How do you model music? (Score 1)

Since the present discussion includes issues about what is good/bad/possible/etc., this leads to the question of whether there is a way to *model* music - to shed light on issues like these, in a systematic way.

Our OMSModel project has been a very successful way to model music - individual works, and individual performances. OMS doesn't look at notes, rather it focuses on how music stimulates or affects a listener. It provides detailed analysis; it can provide numeric ratings (the lowest/highest rankings we have discovered so far are 3/680; the robotic clarinet performance is about a 30). It often provides very good "answers" to the big questions about music (e.g. "What distinguishes good music from bad music?"; "How can music be analyzed in a way that allows for variation in individual responses and preferences?"; "Does music convey meaning?", etc.).

OMS uses ideas from Marvin Minksy's "Society of Mind" (Music is modeled as organized multi-stimulation of a human's "musical agents"), and also some time-honored ideas from academic philosophy. To build out the OMS model (to come up with a good list of "musical agents"), we did in-depth analysis of some great musical works - starting with a couple of Beethoven symphonies and working out from there. Our analysis eventually included a broad range of music: Popular, traditional, world, jazz, classical, avant-garde, and more. We now have a good list of musical agents (somewhat over 100, and fairly stable). It turns out that Beethoven was able to provide over half of the agents; it may sound crazy, but LVB seems to have anticipated some of the ideas in "Society of Mind" (!!)

Detailed materials about OMS are are on www.OMSModel.com

Here are a few things that OMS says about the robotic clarinet performance: The performance rates at about a 30, which ranks it as better than Barney singing his song "I Love You" [18], and ranks it about equal with "It's a Small World" as experienced at Disneyland [31]. It is not up to the level of "The Chimpmunk Song" (Alvin and the Chipmunks, [43]). It is far short of "Baby Beluga" (Raffi, [100]), or a fun pop tune ("Rockin Robin" [120]). These numeric ratings are a rough estimate of the intensity and breadth of the stimulation of the music. Most mainstream pop music is in the range of 100 - 180. Anything 200 and higher is really intense and broad (e.g. the best performances by James Brown or by Miles Davis). 300 and up gets into the range of "mind-blowing" or "unforgettable". Over 300 can be life-changing!

A rating of 30 is a big accomplishment. A large amount of musical and technical analysis would be required to get to this level. To compare: If a child were to pick up a clarinet for the first time and try to play something, this would be about a 3. So, among other things, the Australia team has effectively simulated the result of a lot of clarinet lessons.

The effort to get from 30 up to 120 (= fairly good pop music, good performance) is roughly linear. I.e. if we were to assess the total effort to accomplish the robotic clarinet performance (including prior efforts by other researchers), then multiply by 4 to get the level of effort to get to fairly good pop music. (We haven't estimated the effort to get above 120, but my guess is that the effort starts to become steeply logarithmic at some point shortly after 120. So probably a long long way to get to the level of Richard Stoltzman or Benny Goodman.)