Slashdot Mirror
New Computer Program Determines "Hitability"
illuminatedwax writes "It looks like the process of homogenizing the mediocrity of Top 40 radio is going to be aided by a computer, according to an
article from the Music Industry News Network. Polyphonic HMI has developed a new program called Hit Song Science (HSS) and compares "underlying mathematical patterns" in current hit songs and compares them to a new song to determine if it will become a hit or not. Looks like we can expect even more of the same old junk being recycled for us on the radio, although the article claims that it 'will allow new sounds and styles to flourish.'"
This works if you assume that a "new" or "different" song isn't likely to be a hit.
An NPR article a few years ago reported how music companies decide which Country Music songs will be played on the radio. They cold call people and have them listen to 5 seconds of the song. This tortured person is then asked to rate the song 1-5. The music industry then takes all the songs that get 1s and 5s and discards them. It turns out that often when one group rates a song a 5 another will really hate the song and rate it a 1. So what the industry is really looking for is songs that score 3s.
The reasoning behind all this is that if you hear a song that you'd rate a 1 (hate) you're likely to turn the radio dial. But if you hear a 3 you're not likely to have any particular response at all -- thus you'll stay tuned in for more comercials.
Pop is probably done the exact same way. I guess that's why when you listen to "Classic hits of the [6-9]0s" you hear the same tripe over and again.
Music and Math are closely related, as anyone who's read Godel Escher Bach knows. Musical scores have themes that appear in many different variations such as canons (when a melody is offset in time) and fugues (more complicated than a fugue, read the book if you want to know).
I'm not acoustically talented, and I'm sure I couldn't recognize a fugue or a canon if I heard one, but I know that there is some music that I really like, and that sounds better made and more complete than others. I wouldn't find it hard to believe those songs have properties that a computer could pick out.
For example, have you ever listened to a song for the first time, and been able to anticipate what the next notes would be? I think on some level our brain recognizes patterns that we can't see conciously. With statistical analysis, a program could determine if more hit songs always follow a pattern or a specific pattern (easy to hum songs that get stuck in your head), or if more hit songs would break the melody and hit a note you weren't expecting (like those really mind-blowing high notes).
As a music lover, I would be thrilled if this application worked. It would really enhance websites that try to suggest other songs that you might like based on your favorite songs. In a lot of the music I like, the singer's voice gets deep and gravelly in parts. There could be bands that I hadn't considered listening to who match that profile, and a program like HSS coudl find them.
When this algorithm can understand and reproduce the genius of the Beatles I will be impressed. recognizing tripe is still tripe. Creating tripe is still tripe. Write Yesterday, or In My Life and only then will I retire as a musician.
:)
The thing is, if you listen to a Pop station for long enough, you can easily predict new songs that will be a hit. Without actually enjoying or identifying with the music. My girlfriend naturally likes certain songs, and those songs are always hits (the ones that are over-played for about a year, while the rest are forgotten about).
A good program can easily pick up on the patterns, sure. But these are "hits", eg, songs that fit in with the current political "norm", "style", etc.
A true Hit is something that nobody expected. Something that just became a hit on its own merit. It wasn't non-conformist for the sake of being non-conformist (because it was the style at the time), rather, it expressed some emotion or feeling that the general population was able to identify with. Not because that was the "style" at the time. No computer programmer written by anyone alive today can pick up on that...
But the utter crap that comes out on the "pop" stations today? A perl script (or even a VB script) could pick up on that. It's so obvious. Non-conformist for the sake of being so. Sure, we'll dislike rules, because that's what everyone is doing -- that's what's "in". That's always what is "in".
I almost thing a computer program could pick up on it easier than the average "consumer" would. At least the computer would recognize why it is seeing a song as a "hit", where the consumer just sees a) all the popular radio stations are playing it, and b) all the "cool kids" have that CD, so c) It's cool and I have to like it.
Ugh, I hate the entire radio industry. I hate that radio *is* an industry. I never tune in to radio anymore (currently listening to 3rd Strike - No Light in MP3) because of this crap. It's all about marketabiliy.
If the "average" person likes it, that's fine for them. Me? I can't stand any of the utter shit that pops up on the radio these days. Perhaps I'm not like the "average person". Perhaps I am quite happy this way.
Posting AC because I am drunk
if you had access to the program, and you fed it the songs that were your own personal hits, maybe rated them, it would be better than just about anything else at telling you what else you'd like. Finding you bands you'd never heard of that were actually pretty good. It could allow you to expand your musical horizons rather than forcing you into the narrow spam mold of the cold musical marketing machine. It could easily evolve into a simple web based tool to sell more and a broader variety of music, but they'd never even think of it.
> Rivers uses a mathematical formula when writing
...and on and on
> his songs based on songs by several bands
> including Nirvana. As a huge Weezer fan, I'd have
> to say he's on to something.
Most popular music is almost totally based on formulas e.g.:
- 1-2-3-4, 1-2-3-4, repeat till end
- verse/chorus/verse/chorus/mid 8/chorus
- use I, IV, V, IIm, VIm chords
- sad verses, upbeat choruses (Bruce Springsteen loves this one!)
- something around 120 beats per minute is what gets people tapping their feet in time with the music, even if they're not actually consciously listening to it
- various instruments have their frequency ranges compressed in certain ways; this is what frequently separates the good/big-selling producers from the bad/not-so-big. Listen to multiple albums from the one producer, even across several different artists or styles of music, and you'll pick up the "brand" of specific producers in how they mix specific instruments in the audio spectrum.
Although I'm nowhere near up with state of the art, I'd be surprised if current sound analysis software couldn't detect most/all of the above and spit out some sort of number saying how well a song fits the above "rules".
Finally, if there's any doubt that these formulas exist, check the early 80s bubblegum Brit Pop stuff produced by Stock Aitken Waterman. You could remove the vocals, and what's left of the songs are almost interchangeable.
Exactly. This makes 2 years in a row that an album that was not pushed by the machine made it to number one. Last year's winner was Oh Brother, Where Art Thou, which was good ol' Bluegrass, and this year it was Nora Jones. CNN has an interesting article (considering they are Time/Warner) about the fact that these 2 albums were made hits by word of mouth instead of by radio play.
The commercial music industry is broken. Music is being discovered through word-of-mouth instead of through industry channels. I know that is true for me, I have investigated more music through slashdot posts in the last year than from radio and this means more business for indy (non RIAA) labels. I fact I listen to NPR talk radio on the way to work, and to an '80s stationn if the wife is in the car (she hates talk radio). The RIAA isn't going to sell anything to me this way -- I already have most of the music that the '80s station plays.
Beware of Sleestak