Yonatan Zunger (zunger) wrote,
Yonatan Zunger
zunger

How to identify music...

I really wish there were some systematic way to identify music that you can partially remember.

I've got a bit of music that's been stuck in my head, quite literally, for several years. It's symphonic, with strings leading the melody and some fairly serious horns backing them up. I'm fairly sure it's late 19th- or early 20th-century Russian; it's fairly classical in its style, but has that special bombast of Russian nationalist music. More Tchaikovsky than Rachmaninoff. It sounds like the final movement of a string concerto or (more likely) a symphony, but I'm not sure if it's the main theme or a secondary theme. It's in a minor key -- I think f minor, but I don't really trust my ability to remember an exact pitch after this many years.

Now the question is... given all of this, and the ability to hum the melody (or even transcribe it, I suppose)... how the hell can I figure out what piece it is?

(I've tried going to Amazon and listening to as many samples of pieces as I could find that might match this. Not much luck. There's a lot of music out there.)

Anyone have any ideas?
  • Post a new comment

    Error

    default userpic

    Your reply will be screened

    Your IP address will be recorded 

  • 34 comments
Aren't you the one who works at Google?

(By which I mean: let me know what you find)
Yes. And when I first started, it was this very piece that made me think "Damn, I wish we had a 'hum a few bars and I'll search for it' function..."
There was a thing online a while back, where you could use the space bar to "beat out" a song, and it would come back with results.

But I have no idea what it was called, or if it's still around, or if it has classical type music referenced.
http://www.bored.com/songtapper/

Unfortunately, the classical music coverage is fairly bad. It did correctly identify several Beatles and Police songs that I tapped out.
I've always wanted a search function for that too! Say you could upload a .wav of the melody, can you use it to find an mp3 of the actual song, or would that be a ridiculously difficult audio processing problem?

(And I didn't know you worked at Google. I thought you were a professor at Stanford)
Difficult but not impossible... although I suspect that the legal problem would be far more challenging than the engineering one.

And I left academia a couple of years ago. Nowadays I work on everything from AI/NLP stuff to heavy systems infrastructure. :)

xaosenkosmos

7 years ago

I'm afraid I can't help you with that one -- however, if you find a good system, be sure to let me know. I've been listening to 102.1 lately, and I keep hearing snipets of music that I love, but I have no idea who composed it or who performed it. And it's not like I can look up the lyrics or anything....
That one, I can help you with -- go to kdfc.com, and they have a list of everything they played and when.
Parsons code was developed for just this sort of thing. Musipedia, which I haven't tried myself, uses it among other methods.
Gah! This is so close, and yet so far... It's really hard to input music into their system. I'm going to play with it more later to see if I can get it to work.
Perhaps you heard it in a dream and it exists only in your head...? That's certainly happened to me with books and poems...
No, it's definitely real. I can't compose this well in my head. :)
I've sent your question to my friend writerspleasure, who boasts both a phenomenal memory and a huge backlog of classical music knowledge (Lizt, Chopin, etc).

In the meantime, how much of it do you think you can transcribe? Or you can play the melody or theme on your piano and post it to YouTube for even more help.

zunger

7 years ago

savorie

7 years ago

Phone the head of your local symphony and hum it to him or her. :)

jonquil

November 20 2008, 21:23:02 UTC 7 years ago Edited:  November 20 2008, 21:23:58 UTC

Come talk to omalley. I bet he can help. (He's got an MFA in lute; he's also done AI work in music recognition.) You can wave to me at the same time.
my guess, and it's a pure guess, is that it might be the finale of an orchestration of mussgorsky's pictures at an exhibition: http://en.wikipedia.org/wiki/Pictures_at_an_Exhibition (not very good performance there, but you can check it).

whether it is or is not, you might be interested in the explosive virtuosity of horowitz in his transcription:


Alas, that isn't it... but what a wonderful performance by Horowitz!

writerspleasure

7 years ago

zunger

7 years ago

writerspleasure

7 years ago

susancalvin

7 years ago

susancalvin

7 years ago

zunger

7 years ago

http://www.midomi.com/ is supposedly a service where you can sing or hum a piece and get help identifying it. Not sure if it extends to classical.

Oh, check this out! http://en.wikipedia.org/wiki/Query_by_humming

A list of musical information retrieval systems (probably out of date): http://mirsystems.info/index.php?id=mirsystems
Whoa, how cool!

Hey Yony, hum it to me tonight? I doubt I'll know it, but hey, you never know. :)
You might consider a human search engine such as ChaCha or Mahalo. I don't know if either of these accept audio files as part of the question, but if they do, that might get a fair number of ears and brains on the problem for little cost/time/effort on your part.
I have no useful suggestions. However, you could flesh out the music by making an MTurk task that plays a midi/wav of the snippet you remember, then asking users to select what note should come next. Take the best results for the first note(s), automatically rebuild the midi, and repeat.
I was playing around with some prototype code for doing stuff like that at Google, but never got too far. Pity, really, it was kind of fun.
Query by humming (QBH) has been hit and miss for me --- Midomi (as mentioned by a fellow friend previously) on the iPhone works amazingly well for Meg, but is an utter disaster for me (perhaps due to the fact that I can't carry a tune, even with a bucket...)

There are a couple ACM papers on it from a few years ago, IIRC --- they pointed to other vendors.


Yeah... in the past day I've tried a bunch of QbH engines, and have had pretty bad luck. There's definitely a speech-recognition--like problem here; I can whistle something which sounds tonally fine to a human ear, but which picks up all sorts of bogus stray notes when a computer tries to convert it.