Using Computers for Sophisticated Music Analysis 97
Tom Avril writes "Need an accompaniment for your melody? Seeking a virtual dancer to try out your new choreography? Or perhaps you're making a new TV commercial, and you need a snippet of music that sounds something like Radiohead, but a bit more mellow. Increasingly, sophisticated software can help with these sorts of tasks. We got a look at the latest from the nascent field of Music Information Retrieval, after its conference in Philadelphia: 'A key part of the conference each year is the announcement of results from a sort of software shoot-out — a competition in which various universities pit their music-analysis algorithms against one another. Entrants from more than a dozen countries competed in 18 tasks, using their computers to "listen" to selections of music, then identify such things as the genre, mood, composer or title. The eventual goal: to help people search for music they might like by combing through millions of audio files in a database. ... In another task, the computer had to identify tunes that someone hummed. "The idea is, you go into the karaoke bar and start humming, and the computer retrieves your song," Downie said.'"
Results (Score:5, Funny)
Song: Hit Me Baby
Rating: Shit
Conclusion: Humans are weak and stupid
Action Plan: Terminate Britney Spears
Re: (Score:2)
So why don't you take your self-inspired generalizations and go fuck yourself
Sounds like somebody's generalizations were a little too close for comfort...
Re: (Score:1)
I'm getting regular sex from a girl that looks enough like Tina Fey that both my nerd cred and erection are rock hard and unquestionable.
"Regular" as in "every month" as in "she costs a month's wages".
you will never get within sniffing distance any quality pussy
Whereas you are apparently quite the charmer :)
Re: (Score:2)
Input from programmer :
>>action plan invalid --reason="cannot terminate artist"
>>new action plan
Action Plan 2: Self Terminate
See the good side of it (Score:2)
At least, this would require some melodic humming skill in order to be able to operate the Karaoke machine in the bar.
It will come as a form of protection against all those horrible "I can't sing so I just approximately squeal something completely out-of-tune in the microphone" experiences at the karaoke bar.
Re: (Score:2)
Fixed that for you
The real idea... (Score:5, Funny)
You start humming and the RIAA deducts the money from your account for your reproduction.
Re: (Score:1, Offtopic)
Well, there is some humming involved when approached by the RIAA, but not the kind you think.
Re: (Score:2)
That depends on whether you've had plastic surgery since the last time you paid to hum that tune.
Re: (Score:3, Funny)
I don't think the RIAA has stooped quite as low yet as to interfere with your ability to reproduce
Re: (Score:1)
'Fake Plastic Trees' for the masses (Score:3, Funny)
MORE mellow than 'Fake Plastic Trees?'
This actually may work, especially if you are selling some sort of sleeping aid or anti-anxiety medication.
Re: (Score:1)
"Fake Plastic Trees" does happen to have a 90s-style quiet-to-loud transition into a section with lots of crash cymbals and overdrive and warbly Jonny Greenwood leads. The 'Head have done mellower, including "Bulletproof", off of the same record.
Neat stuff. (Score:4, Funny)
On the one hand, I'd love it if my home stereo could determine what song I was humming and start playing along.
On the other hand, my family would kill me.
Re: (Score:3, Funny)
That reminded me of a funny Married with Children episode [youtube.com]. Oh, Al...
Re: (Score:1)
Better than cloning Simon Cowell (Score:2)
Well it was this or clone Simon Cowell so they can finally start that 24x7 American Idol channel....
Not new tech (Score:4, Informative)
Music is math, but math is not necessarily music. Much of the computerized music based on mathematics alone sounds like atonal shit. Mathematical algorithms can be great for accompaniment but are not meant to replace a human composer.
Re:Not new tech (Score:5, Informative)
"Simply" data mining for music is a significant problem. What data do you mine? The audio signal does not contain all of the perceptual cues we understand as humans, and so things like "rhythm" and "tempo"; i.e. the things in music that get us to dance or tap our feet to it, are hard to pinpoint and even harder to extract.
Other problems, such as the Query-by-humming problem, are further complicated by two intractable problems: 1. People can't sing well out of their head, and 2. What they do sing may or may not bear any resemblance to the actual song they're remembering.
This research uses the latest advances in signal processing, machine learning, psychoacoustics, computer vision and pattern recognition. To compare it to a midi to wave converter is like comparing a paper airplane to the space shuttle.
Re: (Score:2, Interesting)
Re:Not new tech (Score:5, Insightful)
music is not math, man. sorry.
"Music is a communicative signal
comprised of patterns whose
performance and perception are
governed by combinatorial rules, or
a sort of musical grammar" -John Sloboda
that being said, music is not language either, and the brain is not a computer.
if you don't like algorithmic music, or atonal music, don't listen to it. and certainly don't rely on a computer to tell you what art you should absorb and experience...
and, i have to say, i appreciate the reverence for human composers, but mathematics has no bearing on my music apart from the incidental involvement of things like acoustics. to say otherwise is to turn the human composer into a mere algorithm machine.
Re: (Score:2)
Also, the brain *is* a computer, just not one that's terribly similar to our current digital computers. It is a massively parallel pattern recognition system (or can be viewed as such). Theoretically, anything it can do could be simulated
Re: (Score:2, Insightful)
guess we're just gonna have to agree to disagree, man.
you are speaking in analogies to try to define something that is, in itself, abstract. i actually prefer to tell my students that music is "sound organized in time", it's a much better definition.
by the same token i could tell you that language is a specific form of music, since they share certain features, but language communicates specific semantic concepts whereas music does not (in the absence of lyrics).
the difference between art and science is the
Re: (Score:2)
Music is, to a large extent, applied mathematics
In the same way that sex is applied biology? 'Cause that would really be missing the point of sex.
Re: (Score:2)
There is a lot more to music than 'simple math', if you think it's that simple and you have a math degree I urge you to pick up a violin and to try to put your ideas in to practice.
Music is all about expression, the composition is important, and has some mathematical connections but there is really a lot more to it than that.
Oversimplification is hardly ever apt, in music it is throwing out almost all that makes music rich and interesting.
Re: (Score:1)
So you don't use scales? Or synthesizers?
This is a fallacy based on the fact that most perceive mathematics to be cold, unemotional and icky, and that there is a magical spark somewhere machines can't ever emulate.
I think Ray Kurzweil already proved this wrong in the 80s when people had to pick a composition (out of three
Re: (Score:2)
So, Ray Kurzweil proved himself to be a lousy composer, so what ?
Re: (Score:1)
I think live music is different. Only a few people appreciate music just for its mathematical beauty. They are overly represented on slashdot. Most people like music for the emotions it gives them. As a musician, "can you make folks feel what you feel inside" with yo
Re: (Score:1)
In my talk, I will describe five properties that help make music sound tonal â" or âoegood,â to most listeners. I will then show that combining these properties is mathematically non-trivial, with the consequence that space of possible tonal musics is severely constrained. This leads me to construct higher-dimensional geometrical representations of musical structure, in which it is clear how the various properties can be combined. Finally, I will show that Western music combines these five properties at two different temporal levels: the immediate level of the chord, and the long-term level of the scale. The resulting music is hierarchically self-similar, exploiting the same basic procedures on two different time scales. In fact, one and the same twisted cubic lattice describes the musical relationships among common chords and scales.
In fact, Western music is very algorithmic.
Re: (Score:1)
Music is math
That's what I thought when I started my PhD in MIR. I haven't finished it yet :)
Re: (Score:1)
While this technology is very neat, programs which convert sound(wav/mp3) into Midi data have existed for many years.
Incorrect. Bullshitters claiming to convert sound to MIDI have existed for many years. Download that thing and give it a real song (not a flute solo.) The technology required to do it right is probably further off than a Turing test winner. This is one of those things that everyone thinks exists, but it doesn't. On a related note: ask ten random people if they think NASA has an "antigravity room."
Recognising tunes from a simple rendition (Score:5, Informative)
The tune recognition task is easier than it sounds (ha). In fact it's enough to hum the *contour* of the music, i.e. whether it simply goes up or down, for a couple of bars, ignoring the rythm even.
This way of indexing and recognising music is called the Parson Code [wikipedia.org] and is quite effective.
Re: (Score:2, Interesting)
I'm not a Luddite. But...
Is there some value to being able to recall a song, or at least to using your brain to perform the exercise of recalling from memory? This can effectively replace our need to perform this task. (New iPhone app: Hum into it and it will ID the song!) Extend that to all such tasks, which we generally regard as getting in the way of us doing what we really need to.
Example: Calculators very effectively replaced log tables, and we are all grateful for that. But they have also replaced val
Re: (Score:3, Informative)
The iPhone app is already done, Midomi.
http://www.topiphoneresource.info/midomi-iphone-app-sing-hum-or-whistle-to-find-music/ [topiphoneresource.info]
Re: (Score:1)
Re: (Score:2)
In England there exists - or used to exist - a service that you could call with your cellphone, hold it up for a couple of seconds and it would text back a performer / title combo.
Re: (Score:3, Insightful)
We may be creating technology which will gradually make us a non-contemplative people, living only in the moment. And if you live in the moment, you forget the past, allowing those in control to make you repeat it.
Your quote shows that we are not... as I assume you are pretty well immersed in technology since your posting on slashdot.
I would argue that technology has actually made us a more contemplative people.
Visit a farm someday, someplace rural and 'backward', and you will find that as a whole the people there are much simpler and much less likely to question the nature of our existence.
On the flip side, some of the most profound things I have heard said in recent times have come from the mouths of some text mess
Re: (Score:3, Interesting)
One of my senior projects for college used a very similar but more detailed schema in recognizing musical patterns.
In musical terms, a step is the amount of change from note-to-note. The Parson Code is limited in which it only indicates the direction of the pitch, and not the amount. I simply took account the actual half-steps used between each pitch. Like the Parson Code, it would ignore the rhythm, and easily account for identical melodies that are in different keys.
Minuet in G would look something like t
Re: (Score:3, Interesting)
Minuet in G would look something like this:
-7 2 2 1 2 -7 0 9 -4 2 2 2 1 -12 0 5 2 -2 -1 -2 2 1 -1 -2 -2 -1 1 2 2 -4 4 -2
Anyone in particular's Minuet in G (Bach? Mozart? Beethoven? Me?), or all of them? And I thought the key didn't matter, so that would be all minuets?
Re: (Score:2)
Dr. Gallaher?
Re: (Score:3, Informative)
Great Idea (Score:5, Funny)
Self-regulating karaoke. If the computer can't tell what the hell you're singing it's probably best for you to stay off the stage.
-Peter
auto-karaoke (Score:2)
That's the ticket, an auto-karaoke machine!
me: "We're no strangers to looooove..."
machine: [crackles, emits blue smoke]
Re: (Score:2)
that's ok, wait for the version where the industrial laser turret swings around to aim for you ;)
I was about to say... (Score:1)
Re: (Score:1)
Now some sort of alternate front end to the MGP database would solve the problem outlined by TFA.
Re: (Score:1)
Re: (Score:2, Interesting)
the song features pop rock qualities, folk influences, a subtle use of vocal harmony, use of string ensemble, major key tonality, a vocal-centric aesthetic, a good dose of acoustic guitar pickin' (sic) a dynamic male vocalist, electric pianos, acoustic rhythm guitars, romantic lyrics...
While a sophisticated computer may be able to detect some of these characteristics, I stand by my comment.
Re: (Score:3, Insightful)
The Music Genome Project is definitely tracking things that (at this point) take a human to notice. Features like "great trumpet solo" or "ambiguous lyrics" are quite a bit beyond the sorts of musical features being extracted by the tools described in the papers at the conference, based on the few I looked at.
Humans are fantastic musical processors. Computers not so much. Which is what makes the problem so fascinating, I think.
Old. (Score:2)
Advanced music analyzing isn't new. There's software out there that can already successfully identify instruments, chords, progressions and even whole musical concepts, so classifying them by genre, intensiveness, instrument set and the like is just a querying problem.
Now, GENERATING music with math - that's where the fun begins. I've seen lots of approaches already, most of the successful ones based on fractals and powers-of-two, though I admit that they work only because of certain hardcoded "musical idea
Re: (Score:1)
unnecessary (Score:5, Funny)
You don't need software algorithms for that, just go download a Coldplay album. Except maybe replace "mellow" with "soulless."
Amarok Plugin (Score:1)
About this Karaoke++ (Score:1)
genre
Classical - lots of orchestral instruments
Hip-hop - repetitive loops of drum / bass with melodic lyrics
Rap - repititive loops of drum / bass with unmelodic lyrics
Techno - Synthetic (simple wavelength) instruments
etc.
mood
Major and minor chords.
composer or title
Huge music database + measuring contours, with "almost match" working. This doesn't sound incredibly sophisticated to me.
Re: (Score:2)
To help people search for music they might like ? (Score:3, Insightful)
The problem with this technology, and what Pandora or LastFm applied, is that the programs tend to choose always the same kind of music, and it's boring.
When I listen to music, I like to have some variety, not always playing the same thing, again and again, in different forms.
I like listening to one genre, and then switch to another genre, and the programs are unable to provide that.
Re: (Score:2)
It's pretty easy to get a mix with Pandora. Either seed a "station" with songs representing different styles, or create multiple stations and then listen to the "QuickMix".
Re: (Score:1)
I like listening to one genre, and then switch to another genre, and the programs are unable to provide that.
Not so, Arthur Flexer and co. had a great paper [ismir.net] at the conference in question on generating playlists with a specified start and end songs. The system attempts to make smooth transitions between songs that take you through N tracks from the track you started with (in one genre) to your end song (potentially in another genre), passing through whatever is in between. I believe its a publicly available on an http://fm4.orf.at/soundpark [slashdot.org]">Austrian radio site. Paul Lamere [sun.com] at Sun Labs (on the 'Search inside th [sun.com]
Re: (Score:1)
Just a quicker way to find the actual results to the 2008 MIREX task sets:
MIREX 2008 Results [music-ir.org]
And the one page pdf summary: Results Summary PDF [music-ir.org]
Re:To help people search for music they might like (Score:1)
Virtual Dancer (Score:1)
hmmm... (Score:1)
i can tell genre, mood, composer/artist, and title by looking at the sheet music, and i can probably do it faster than a computer.
this isn't analysis, it's cataloging based on surface features. if we're talking about popular music, is the computer going to do text recognition on the lyrics so when i feel sad i can ask for a song about somebody that has it worse than me?
seems a bit shallow to me. it won't change the fact that despite the surface features, there is good music and bad music. i'd like to see
Re: (Score:1)
Reminds me of that guy who could tell you which classical music piece was stored on a record just by recognising the patterns in the grooves.
Wow,..Clearchannel must be crying with joy... (Score:1)
Sofistokated (Score:2)
"Need an accompaniment for your melody? Seeking a virtual dancer to try out your new choreography? Or perhaps you're making a new TV commercial, and you need a snippet of music that sounds something like Radiohead, but a bit more mellow."
To paraphrase Douglas Adams: This is apparently a use of the word of 'sophisticated' of which I was unaware. And here I thought things like time/frequency analysis of notes and harmonics of chord sequences using continuous wavelet transform or analysis of dimensional comple
Re: (Score:1)
Re: (Score:1)
The same people who don't want you to share your music with anyone are the ones supporting research in getting computers to pick your music for you, or manipulate music generation algorithms to alter song 'recipes' to produce that which they want you to find more acceptable.
That maybe the goal of some researchers, but many of us that work on audio music recommendation were and are excited about the prospect of finding long tail music or creative commons music. There's nothing evil about recommending a similar track to what you're listening to or like - particularly if it comes from a small artist trying to make his or her independent way in the world. In fact how does that benefit groups of record labels climbing over themselves to sue you for downloading their music illegally
And this is news how ? (Score:2)
Re: (Score:2)
actually afaik Pandora is an example of an "expert-driven" service. Songs were rated by experts on various categories and this metadata was used to organize them. Probably some automated analysis was also used to speed up the process.
This is different from LastFM which uses user-generated metadata, which is a concept that scales way better but may be less accurate.
Both of these can be mixed with automated techniques to enhance the results of course.
Re: (Score:1)
For more reading about ISMIR ... (Score:2)
From a researcher at Last.fm: http://mir-research.blogspot.com/2008/09/ismir-2008-demos.html [blogspot.com]
A researcher from strands: http://www.scwn.net/2008/09/ismir-past-present-and-future/ [scwn.net]
And lots of posts from my blog:
Could Help Me (Score:1)
Re: (Score:2)
hum it, post it as an mp3 and ask the world what it is, maybe that would get you an answer ?
Wow, would that actually help! (Score:1)
Their response? "We like the intro from the first song, and the melody [read "theme" or "advent" or some other relevant term] from the second song. We'd like them both, thanks. Please make it so."
Does it matter that the key's of the songs aren't the same? Does it matter that the tempos don't match? Or tha
It needs more cowbel! !! (Score:2)
Definitely!