Halfway down are two little 'click to play' boxes that claim to be Bach like. I'm no virtuoso, but I can play the piano a little bit and that stuff sounds so mechanistic to me it's not even funny. The first impression you get is of a very bad player piano roll (the caption says the 'tinny' sound is because of the MIDI, but that's not what I'm getting at, it's the totally 'dead' delivery, like a corpse playing the piano). And that's just the first bit, the compositions are absolutely boring, even the simplest piece from 'das Notenbuchlein' blows this away.
Glenn Gould and Dinu Lipatti are turning over in their respective graves, Bach himself is positively spinning.
If 'most people' really can't tell the difference between that music and the real thing it is sad.
I don't doubt that in the future we'll reach a point where computers will compose music on par with human composers but these two samples leave me underwhelmed. I think the proper thing to do here is to appreciate the fact that the pig can dance at all but to my ears it is just painful.
I seem to remember a while ago there was a program featured on - maybe Slate - that had a much more impressive output, but I think it was actually played by a human and just written by a computer.
That would at least get away from the 'player piano' sound, but it wouldn't fix the problems with the composition.
In Amsterdam there is a small, and relatively unknown museum full of pianolas (player pianos) run by Kasper Janse. I know the guy because I once found a pianola somewhere and it ended up in his museum. When I delivered it there he took me through the rooms and showed me his machines, one by one and I told him I could close my eyes and pick out any pianola compared to the 'real' thing (a person playing).
So he showed me his 'piece the resistance'. A one-off, a machine constructed to be the 'state-of-the-art' in modern recording technology. It featured a roll played by Rachmaninoff himself, with as many parameters captured as possible, including stroke velocity, pedals, the works, and at a ridiculously high sample rate (for the day, at least).
The effect was amazing, it was as though the ghost of Rachmaninoff had come to life and even while writing this I still get gooseflesh. It was the most amazing thing I've ever seen in music, and it was lightyears ahead of the compact discs of today in terms of delivery. What an amazing sound. And, if you've never seen a 'player piano' at work, the keys move as though invisible fingers press them down. The reason there is only one of those machines (and very very few rolls, they had to be 'recorded' by virtuoso players instead of 'punched') is that the phonograph happened.
Conlon Nancarrow (http://en.wikipedia.org/wiki/Conlon_Nancarrow) did some really wild stuff with similar pianos (technically, I think they're called reproducing instead of player). His stuff is so complex that it's impossible to humanly play, so punching paper was his only option. I'd certainly consider him a hacker in that regard.
The audio is converted from MIDI files, hence the tinny sound of the piano.
It's not the sound itself that is the main point here. It's the composition. MIDI files describe only notes and additional attributes assigned to them - ways of accenting and going from one to the other. You could turn it back into a standard score and play it yourself really.
I'm really surprised the author of the article didn't try to get a snippet to play (edit: they do on the page 2 of the article). The music still doesn't quite do it for me but if you consider the potential of future versions it is pretty amazing.
For a fairer comparison, on the second page of the article there are some clips that sound much less wretched.
It's hard to be sure whether a piece of music is any good on the basis of a sample less than a minute long, but FWIW I'd say they're fantastically impressive as productions of a computer program and distinctly unimpressive as serious musical compositions.
They sound mostly like 'variations', not really compositions. I'm not sure if I can make the distinction clear because I'm not a composer and the difference is subtle, but a variation on a theme will not be the same as the theme but is a fairly logical derivative of one, whereas a composition takes place on a 'blank slate'.
Now, of course this program - just like a real composer - works against a backdrop of the knowledge available about music of a given era. But when Bach wrote his first fugue it was a breakthrough, and even if after that each and every other fugue was linked back to that first one in a more or less direct way (some sound like close cousins, some are much further apart and some are different pieces altogether) then the act of creating such a variation is a composition of sorts.
But when a composer speaks of a 'variation', he does so in a subtly different way than an ordinary person would, to a composer a variation is 'obvious' given the first few bars of the variation and the original piece.
And that's what I'm getting at when I say these sound to me like variations. They're along an existing or at least familiar theme and they've changed. It's as though sampling has met classical music and it's bits of everything but not a 'whole' by itself.
This is very tricky to put in to words.
For a computer program though, it's definitely quite a step forward compared to the usual. Some of those pieces on the second page could have you fooled for a few seconds. The fact that some more modern composers make ear-splitting stuff doesn't help either because this computer program seems to be at least well ahead of some live humans :)
> But when Bach wrote his first fugue it was a breakthrough, and even if after that each and every other fugue was linked back to that first one [...]
What makes you say that? Bach did not invent fugues, so why should his first fugue be a breakthrough--compared to other fugues--and why should every fugue henceforth go back to Bach's first fugue? (See http://en.wikipedia.org/wiki/Fugue#History)
I'm sorry, complete goof on my part, I was not aware of any fugues before the baroque period, and I saw Bach as their main proponent and originator, which was apparently completely wrong.
As for the reference back, I meant to the principle of a fugue, as an original form.
And not that original :)
The wikipedia piece gives me a little bit of cover, it says "It was in the Baroque period that the writing of fugues became central to composition", but then goes on to list a whole slew of composers that got there before Bach did. So it's not even a fig-leaf.
Johann Sebastian Bach was even viewed as conservative and old-fashioned in his time. He was more well known as a teacher of music, but not as a great composer.
I play the piano in general, and have played some Bach in particular, and I'm a bit confused by your reaction.
Are you referring to the two tracks on the bottom of the first page? If so, I don't see text that claims that they are Bach-like. The paragraph that follows those two tracks mentions Bach-like compositions, but the click-to-play boxes corresponding to that text are on the second page.
In particular, the two click-to-play tracks on the second page called "Fugue" are the two supposedly Bach-like tracks.
If, instead, you were referring to those two tracks, I'm wondering: do you have much experience with Bach's fugues in particular? When you strip away the interpretation that a live performer brings to Bach fugues, what is left is indeed very mechanistic, because the construction of the composition is largely based on starting with a theme, transforming it under various rules, and playing those transformations against each other. With interpretation they can be sublime. Rendering a satisfying interpretation of Contrapunctus I is among the most joyful experiences I've ever had at the keyboard.
But not all of Bach's fugues reach such great heights. There are 24 of them in each of Well-Tempered Clavier Books I & II — and there are some more mundane fugues within that are closer to the two fugues on page 2 in terms of compositional brilliance.
> I play the piano in general, and have played some Bach in particular,
Neat, colour me positively jealous, the only piece I can play (by Bach) is the first of the WTK and the opening bars of partita I (which I'm insanely proud of, it took me a very long time to get that right, but as soon as the left hand starts to carry the melody I'm totally lost).
So, I'm not a pianist by any definition, just to get that cleared up :)
> Are you referring to the two tracks on the bottom of the first page?
On that page it says:
"Cope has been writing software to help him compose music for 30 years, and he long ago reached the point where most people can't tell the difference between real Bach and the Bach-like compositions his computer can produce."
So I took those two - and the rest - to be samples of what it is to be 'Bach like', and while I can see some similarities (at least, more than with other composers, for those first two and the fugues) the pieces are terribly simple.
I know about the fugue rules, for instance, one way to condense the score was to give just one or two voices and let the player figure out what the other ones should be since the transformation is an entirely logical one (I hope I understood that correct), not that I could do it, but I think I get the principle, it is somewhat analogous to the development of a mathematical series.
The more mundane of the fugues, while not on par with some of the more spirited ones, in my opinion can compete easily with the samples of the two fugues on the second page, they are as I tried to put down elsewhere (but probably failed) to me like thin variations, and then poorly executed.
I'm a bad judge of 'compositional brilliance', but I think that when Bach wrote his fugues he tried to express some idea, it has a feeling of 'going somewhere' and this stuff could probably be stretched endlessly without going anywhere, it sounds like phrases ripped from a book at random strung together. Does that make any sense to you?
That doesn't mean that it isn't an impressive feat of programming though, and if he keeps at it there is a good chance that within a few years it will be at a level where it can truly compete. But it does not give me that feeling just yet.
I can't play worth shit, but I've listened to music since before I could talk, my dad was totally fanatical about classical music, it took me until I was 13 or so before I heard my first popular music (and I never stopped listening to that either), there is something very impressive about this music in sense that it just 'could be' composed by a person, but that - at least, we have the say-so of the artist - it was composed entirely by machine and that it becomes hard to tell the difference.
But just like a chess computer, it looks like intelligence is behind it and is making the decisions, but in the end it was just 'brute force', so I would not expect this software to suddenly come up with a new style or an undiscovered branch of music.
And that's where the real genius in composition lies.
Again, I'm totally jealous of your skills, not because I don't realize that you must have pumped thousands of hours in to learning to play like that, but simply because if I had not fallen in love with programming I would have stuck to music.
That makes sense now. I saw that text, but thought it was too far from those two files to indicate any connection.
if I had not fallen in love with programming I would have stuck to music.
I hear ya. There's a definite connection between programming & music (and/or mathematics and music) isn't there? I wish I had more time for music. I haven't touched it for half a decade now, and I'd really have to work to get even some of it back.
I haven't yet decided what I think of this Emily Howell thing. I think I'd like to hear an entire composition interpreted by a real player. These snippets don't really give enough of a dramatic arc so it doesn't seem like a fair way to compare them. To me a lot of the real Bach fugues don't really come together until the last few bars, and I wonder if any of these real pieces would be the same way.
> one way to condense the score was to give just one or two voices and let the player figure out what the other ones should be since the transformation is an entirely logical one
Fugue construction isn't mechanical; if you had only one or two voices you couldn't reliably work out all the others. (You might be able to make a plausible guess.)
No sense of timing or rhythm. Even an algorithm that adds a random delay to notes would make it sound more natural. I listened to previews of the album on iTunes and it sounds like crap.
Not yet, but I might set aside some time to do it. I'm impressed with the snippets, but I have a hunch that I'd be able to tell a generated work from the real stuff.
No it isn't. What an absolutely ridiculous claim. The two "Bach-like" pieces are rubbish, full of strange discontinuities and random meanderings, not to mention the rhythm just hammering away in a very set pattern. Reminds me of terrible beginning piano workbook pieces, only less musical.
On the next page, the other "fugue" does bizarre things that I don't think a human composer would. The beginning is entirely discontinuous, and doesn't really sound like a beginning. A real fugue isn't just something being played repeatedly in different intervals.
If that's the best that the program has to offer, then I'm not sure what it's accomplished. Just because it can combine complicated things doesn't mean its composed anything, because it doesn't understand the parts. If someone copied and pasted parts of great books, and then tried to weave the parts together, would the result be any good? Or even great books by the same author. Maybe that's a weak analogy though.
I feel sorry for the Slate writer if he can't tell the difference between the Bach and this noise, he's seriously missing out.
Ah! Discontinuous, that nails it perfectly. I couldn't find the right word, but that's exactly the feeling, it breaks up. The best I could come up with was 'random sentences from a book'.
This is music sort of like the writing you get from a Markov text generator. Except much better. Has anyone gotten any good thoughts or ideas from a Markov text generator? And that illuminates something about music: the patterns aren't as complex as language.
This is going to cause a giant re-ranking of composers, as those whose music seems most like computer-generated stuff will seem less impressive now. (That will be wrong, but it's what people will think anyway.) Interestingly, the ones who create the noisiest stuff will benefit the most, whereas those whose compositions are perfectly expressed in the linear notation of sheet music will lose out. Ramones in, Beethoven down, Rachmaninoff out.
The MIDI samples were a really poor choice to introduce the program's compositions, but those snippets do sound a lot like Bach chorales stylistically, perhaps at the level a competent (if uninspiring) student of harmony.
The chorales exist online in machine-readable formats at http://www.jsbchorales.net/ (a great resource for doing comprehensive analysis on chorale harmony -- wouldn't doubt if the same data served as input for the Emmy program).
This technology would be great for amateur filmmakers, especially those who aren't skilled musicians. Imagine a UI with sliders to create the desired mood, tempo, emotion, etc. Maybe the result isn't a John Williams score, but it sounds professional, is easy to produce, and is royalty free.
I suppose the drawback is if the software is easily accessible, and everyone uses similar databases of sources, the compositions could all start to sound alike.
37 comments
[ 3.1 ms ] story [ 93.0 ms ] threadGlenn Gould and Dinu Lipatti are turning over in their respective graves, Bach himself is positively spinning.
If 'most people' really can't tell the difference between that music and the real thing it is sad.
I don't doubt that in the future we'll reach a point where computers will compose music on par with human composers but these two samples leave me underwhelmed. I think the proper thing to do here is to appreciate the fact that the pig can dance at all but to my ears it is just painful.
In Amsterdam there is a small, and relatively unknown museum full of pianolas (player pianos) run by Kasper Janse. I know the guy because I once found a pianola somewhere and it ended up in his museum. When I delivered it there he took me through the rooms and showed me his machines, one by one and I told him I could close my eyes and pick out any pianola compared to the 'real' thing (a person playing).
So he showed me his 'piece the resistance'. A one-off, a machine constructed to be the 'state-of-the-art' in modern recording technology. It featured a roll played by Rachmaninoff himself, with as many parameters captured as possible, including stroke velocity, pedals, the works, and at a ridiculously high sample rate (for the day, at least).
The effect was amazing, it was as though the ghost of Rachmaninoff had come to life and even while writing this I still get gooseflesh. It was the most amazing thing I've ever seen in music, and it was lightyears ahead of the compact discs of today in terms of delivery. What an amazing sound. And, if you've never seen a 'player piano' at work, the keys move as though invisible fingers press them down. The reason there is only one of those machines (and very very few rolls, they had to be 'recorded' by virtuoso players instead of 'punched') is that the phonograph happened.
If you're ever in Amsterdam, well worth a visit:
http://www.pianola.nl/Pianola_Museum/Homepage.html
http://www.youtube.com/watch?v=muCPjK4nGY4
eerie!
You're correct about the 'reproducing pianos', I didn't know the correct English term.
The audio is converted from MIDI files, hence the tinny sound of the piano.
It's not the sound itself that is the main point here. It's the composition. MIDI files describe only notes and additional attributes assigned to them - ways of accenting and going from one to the other. You could turn it back into a standard score and play it yourself really.
I realize the 'delivery' is the midi, but even if the compositions were played by a human, so we'd rule that part out, they would still be boring.
A real Bach piece played on a midi synth sounds lightyears ahead of this.
http://www.youtube.com/watch?v=QEjdiE0AoCU&feature=relat...
I'm really surprised the author of the article didn't try to get a snippet to play (edit: they do on the page 2 of the article). The music still doesn't quite do it for me but if you consider the potential of future versions it is pretty amazing.
The wind in the planes crossed with something undefinable.
http://www.youtube.com/watch?v=1N5fvYejvXk
It's hard to be sure whether a piece of music is any good on the basis of a sample less than a minute long, but FWIW I'd say they're fantastically impressive as productions of a computer program and distinctly unimpressive as serious musical compositions.
Now, of course this program - just like a real composer - works against a backdrop of the knowledge available about music of a given era. But when Bach wrote his first fugue it was a breakthrough, and even if after that each and every other fugue was linked back to that first one in a more or less direct way (some sound like close cousins, some are much further apart and some are different pieces altogether) then the act of creating such a variation is a composition of sorts.
But when a composer speaks of a 'variation', he does so in a subtly different way than an ordinary person would, to a composer a variation is 'obvious' given the first few bars of the variation and the original piece.
And that's what I'm getting at when I say these sound to me like variations. They're along an existing or at least familiar theme and they've changed. It's as though sampling has met classical music and it's bits of everything but not a 'whole' by itself.
This is very tricky to put in to words.
For a computer program though, it's definitely quite a step forward compared to the usual. Some of those pieces on the second page could have you fooled for a few seconds. The fact that some more modern composers make ear-splitting stuff doesn't help either because this computer program seems to be at least well ahead of some live humans :)
What makes you say that? Bach did not invent fugues, so why should his first fugue be a breakthrough--compared to other fugues--and why should every fugue henceforth go back to Bach's first fugue? (See http://en.wikipedia.org/wiki/Fugue#History)
As for the reference back, I meant to the principle of a fugue, as an original form.
And not that original :)
The wikipedia piece gives me a little bit of cover, it says "It was in the Baroque period that the writing of fugues became central to composition", but then goes on to list a whole slew of composers that got there before Bach did. So it's not even a fig-leaf.
Thanks for the correction!
I'm sure grateful for all the works that 'mediocre' composer left for us to enjoy.
Are you referring to the two tracks on the bottom of the first page? If so, I don't see text that claims that they are Bach-like. The paragraph that follows those two tracks mentions Bach-like compositions, but the click-to-play boxes corresponding to that text are on the second page.
In particular, the two click-to-play tracks on the second page called "Fugue" are the two supposedly Bach-like tracks.
If, instead, you were referring to those two tracks, I'm wondering: do you have much experience with Bach's fugues in particular? When you strip away the interpretation that a live performer brings to Bach fugues, what is left is indeed very mechanistic, because the construction of the composition is largely based on starting with a theme, transforming it under various rules, and playing those transformations against each other. With interpretation they can be sublime. Rendering a satisfying interpretation of Contrapunctus I is among the most joyful experiences I've ever had at the keyboard.
But not all of Bach's fugues reach such great heights. There are 24 of them in each of Well-Tempered Clavier Books I & II — and there are some more mundane fugues within that are closer to the two fugues on page 2 in terms of compositional brilliance.
Neat, colour me positively jealous, the only piece I can play (by Bach) is the first of the WTK and the opening bars of partita I (which I'm insanely proud of, it took me a very long time to get that right, but as soon as the left hand starts to carry the melody I'm totally lost).
So, I'm not a pianist by any definition, just to get that cleared up :)
> Are you referring to the two tracks on the bottom of the first page?
On that page it says:
"Cope has been writing software to help him compose music for 30 years, and he long ago reached the point where most people can't tell the difference between real Bach and the Bach-like compositions his computer can produce."
So I took those two - and the rest - to be samples of what it is to be 'Bach like', and while I can see some similarities (at least, more than with other composers, for those first two and the fugues) the pieces are terribly simple.
I know about the fugue rules, for instance, one way to condense the score was to give just one or two voices and let the player figure out what the other ones should be since the transformation is an entirely logical one (I hope I understood that correct), not that I could do it, but I think I get the principle, it is somewhat analogous to the development of a mathematical series.
The more mundane of the fugues, while not on par with some of the more spirited ones, in my opinion can compete easily with the samples of the two fugues on the second page, they are as I tried to put down elsewhere (but probably failed) to me like thin variations, and then poorly executed.
I'm a bad judge of 'compositional brilliance', but I think that when Bach wrote his fugues he tried to express some idea, it has a feeling of 'going somewhere' and this stuff could probably be stretched endlessly without going anywhere, it sounds like phrases ripped from a book at random strung together. Does that make any sense to you?
That doesn't mean that it isn't an impressive feat of programming though, and if he keeps at it there is a good chance that within a few years it will be at a level where it can truly compete. But it does not give me that feeling just yet.
I can't play worth shit, but I've listened to music since before I could talk, my dad was totally fanatical about classical music, it took me until I was 13 or so before I heard my first popular music (and I never stopped listening to that either), there is something very impressive about this music in sense that it just 'could be' composed by a person, but that - at least, we have the say-so of the artist - it was composed entirely by machine and that it becomes hard to tell the difference.
But just like a chess computer, it looks like intelligence is behind it and is making the decisions, but in the end it was just 'brute force', so I would not expect this software to suddenly come up with a new style or an undiscovered branch of music.
And that's where the real genius in composition lies.
Again, I'm totally jealous of your skills, not because I don't realize that you must have pumped thousands of hours in to learning to play like that, but simply because if I had not fallen in love with programming I would have stuck to music.
That makes sense now. I saw that text, but thought it was too far from those two files to indicate any connection.
if I had not fallen in love with programming I would have stuck to music.
I hear ya. There's a definite connection between programming & music (and/or mathematics and music) isn't there? I wish I had more time for music. I haven't touched it for half a decade now, and I'd really have to work to get even some of it back.
I haven't yet decided what I think of this Emily Howell thing. I think I'd like to hear an entire composition interpreted by a real player. These snippets don't really give enough of a dramatic arc so it doesn't seem like a fair way to compare them. To me a lot of the real Bach fugues don't really come together until the last few bars, and I wonder if any of these real pieces would be the same way.
Fugue construction isn't mechanical; if you had only one or two voices you couldn't reliably work out all the others. (You might be able to make a plausible guess.)
It just doesn't make sense, as you mention, because it's not producing a performance, it's producing the underlying composition.
This can be read two ways...
On the next page, the other "fugue" does bizarre things that I don't think a human composer would. The beginning is entirely discontinuous, and doesn't really sound like a beginning. A real fugue isn't just something being played repeatedly in different intervals.
If that's the best that the program has to offer, then I'm not sure what it's accomplished. Just because it can combine complicated things doesn't mean its composed anything, because it doesn't understand the parts. If someone copied and pasted parts of great books, and then tried to weave the parts together, would the result be any good? Or even great books by the same author. Maybe that's a weak analogy though.
I feel sorry for the Slate writer if he can't tell the difference between the Bach and this noise, he's seriously missing out.
This is going to cause a giant re-ranking of composers, as those whose music seems most like computer-generated stuff will seem less impressive now. (That will be wrong, but it's what people will think anyway.) Interestingly, the ones who create the noisiest stuff will benefit the most, whereas those whose compositions are perfectly expressed in the linear notation of sheet music will lose out. Ramones in, Beethoven down, Rachmaninoff out.
The chorales exist online in machine-readable formats at http://www.jsbchorales.net/ (a great resource for doing comprehensive analysis on chorale harmony -- wouldn't doubt if the same data served as input for the Emmy program).
I suppose the drawback is if the software is easily accessible, and everyone uses similar databases of sources, the compositions could all start to sound alike.