What is the smallest amount of information that would permit a computer to identify any piece of music, popular or classical?
Is rhythm alone sufficient?
Would this factor include the ability to identify Indian and Chinese pieces of music?
NOTE: I except individual pieces of music—like Aura Lee and Love Me Tender–which are identical.
The title, I suppose.
Rhythm is certainly not sufficient, since there are approximately one billion songs written in 4/4 time.
Time signature is only a small part of rhythm. The actual rhythm of the melody can certainly be used to identify much music. There even was a website that did this (called Song Tapper, but it doesn’t seem to be up anymore.) And, from my recollection, it actually did work pretty well.
You can narrow the search down by specifying merely the relative pitch, or at least you could. There was a site where you would specify no more than something like <note><note><note><note><lower note><lower note><higher note><higher note><higher note> and the site would say “Heart and Soul” and maybe give you a button so you could confirm it. I say “was”, I can’t find it now. Probably closed by the RIAA or some such.
Yeah, I remember that one, as well. It would identify songs by “contour.” I believe it was on this website, but I’m getting 503’ed: http://www.musipedia.org/
Looking at the Google results from Musipedia, it seems that one also did have a tapping option on it, to look up by rhythm, as well as by playing a Java-based piano, humming into your mic, etc.
Not to answer the question regarding a computer doing this, but I recall a radio show where the dj challenged some guests with naming rock songs based on just the first beat of the song. The guests were pretty good. I’m sure this was/is a common dj thing.
I suppose, if the computer stored all the audio tracks, or possibly a score of all the songs in a given range, it would be a small matter of converting the audio to it’s electronic equivalent and do a brute-force matching search, with some pruning to get a genre. Could it be as good as a human? I wouldn’t say no.
I started a thread on this topic 8 years ago. I posted the approximate melody “ccdeca’G’CdecA’ccdEegAagageDcca’DDdcEcCageca’C.” In barely one hour I had my answer:
Thanks again, randomstar !
Clicking I see that that response to me is the and only post that randomstar ever made at SDMB. :eek:
Anyone remember “Name That Tune”? When I was a kid (I’m 73) the computer between my ears would nail the majority of examples on the second or third note. I suspect the examples were mostly drawn from the day’s popular music, and the radio was always on at our house.
Basically, how does the Shazam app work?
This is reportedly a well-known game among the kids at the conservatorium (not limited to rock songs, of course).
There was TV show called ‘Name that Tune’ both in the UK and the US in the 1970s that invited competitors bet on their ability to name a tune based on progressively fewer notes.
https://www.youtube.com/watch?v=kGn5lT8Wo-w
Makes me wonder whether computer programs like Shazam would win such a game.
Pattern recognition is something that human beings are very good at and computers find challenging.
Remember, though, that in the “I can name that tune in 5 notes/4 note/3 notes/etc” section, there is a clue given as to the song, so it’s really most a trivia game at that point. That’s why you have so many of these end up as “I can name that tune in 1 note.” They’re not really naming the tune in one note. They’re naming it because they figured out the answer to the clue.
In the other rounds, it was different. Then it was just a simply playing of the song, and usually contestants would get it by four or five notes. There was also another round where the answers were thematic, so you’d get a hint as to what the title was about.
And “one note” might also be a chord. I know that there have been times that I’ve recognized a song by its opening chord (though of course, that still won’t usually be enough).
By “identify a piece of music,” do we mean matching a sound sample to a specific recording? Or identify a melody, so that it would also recognize a cover by a different artist as the same song? Those are very different computational problems to solve. Shazam can only do the former - it won’t identify a classical music recording unless that exact same recording is in their database.
I’ve done exactly what **Chronos **said, identified a song in a contest with just the opening chord. Looking back, it helps if you can limit the dataset. In this case, the chord was played by a keyboardist in a band performing rock and pop of the last few decades to an American audience. So given a) instrument b) genre c) era and d) geographical origin you could narrow down the selection quite a bit.
It depends on the song, and on the contest rules. One example is “Hard Day’s Night”, which a Beatles fan can positively identify in a tiny fraction of a second - it’s not even necessary to hear all of the first note. But that’s because of the unique sound of the particular instruments playing the first note, and the overall sound of the recording. Change the contest rules so that every example is played with bland computer-generated tones, or even played on brand-new recordings all made by the same skilled-but-bland cover band, and the same example becomes much more difficult.
In short, are you recognizing the song? Or are you recognizing the band, the recording studio, the marginally out-of-tune and very recognizable piano they used that day, and so on?
We played a version of “Name That Tune” in music appreciation class in junior high. There was one song that we all knew before the music even started.
None of you will be surprised to read that it was
“Bennie and the Jets.” We identified it based on the applause.
In Name That Tune that one note was indeed just one note. They only played single note melodies on some kind of keyboard.