The title might be misleading.
Let me give you the backstory.
My wife came home from college and said “Honey !! You’ve got to listen to this…” and started playing this track by this band called Fun: We are Young. ft. Janelle Monae
I muttered something stupid and got back to what I was doing.
Went back to work…the song is playing in my head.
While driving back stopped by at a gas station, it’s still playing.
Had dinner. Watched Colbert and J-Stew. It’s still there..
Fired up Google.
Here is my working set:
1) Can you measure the likability of a song ?
2) Can you measure replayability ? (Malcolm Gladwell in Blink, why Kenna’s music career never took-off because his songs were never picked-up by Radio stations and played over and over)
This has huge potential for music industry if you can figure out some way to measure “re-playability”.
3) In a WSJ Article Michaleen Douclef stated that, Adele employs a musical device called a “appoggiatura”.
An appoggiatura is a type of ornamental note that clashes with the melody just enough to create a dissonant sound. “This generates tension in the listener,” said Martin Guhn, a psychologist at the University of British Columbia who co-wrote a 2007 study on the subject. “When the notes return to the anticipated melody, the tension resolves, and it feels good.”
Are all likeable songs “appoggiatura” ? That’s too narrow a domain. Results might be misleading.
4) Can we split a song into X-variables like pitch, tone and do some analysis to find songs in a particular “zone” ?
(I honestly dont know what these potential variables are..)
5) Thinking like a System Admin:
Can we create a Bayesian Filter like Spam Bayes and train it to spot a good-song from a bad-song?
Lets not take songs in random.
Lets setup the test with Top200/500 in Billboard and evaluate “good mail” from “spam mail”.
We are trying to predict which songs in Top 100-500 range will break in to Top-50/100
If we are wrong, we train the filter.
Can we actually do a Bayesian Filter for mp3’s ?
Let’s see if we get somewhere…