I was pretty excited to try this out (there is an online demo linked), however after some playing around I found the experience lacking. Not that things sounded _bad_ persay but that I found it incompatible with actually playing anything that sounded nice.
I think part of my disatisfaction is due to pattern recognition. As humans we recognize patterns EVERYWHERE, even when they don't exist. So what happens (or what I believe might be happening) is that I think I recognize a pattern, try to play a note to continue it, then the NN decides otherwise. Although the resulting sound may be nice to an outsider it breaks my internal model, making it sound "bad" to me.
I do however like the idea of augmenting insturments to make them easier to play, I'm personally guitly of having many potential songs in my head but not having near enough the skill to execute them.
I think the dissatisfaction you're talking about arises necessarily from projects that generate notes with a network trained on a bunch of MIDI data. Our ears expect music to have structure (chord progressions, etc) but models like this only really mimic small-scale features of the source. So any given phrase of 3-5 notes can sound musical, but over the course of a measure or two the illusion breaks and it sounds largely indistinguishable from a random walk of notes.
That's the analogy I usually make, yeah - it's like generating a story with a neural network trained on a bunch of stories. You'll get grammatical phrases, and sometimes sentences, but naturally none of it will make any sense.
>I do however like the idea of augmenting insturments to make them easier to play, I'm personally guitly of having many potential songs in my head but not having near enough the skill to execute them.
Yeah, we've been doing that for a while with music tech starting from basic sequencers, arpeggiators and looper pedals, down to programmable automatic accompaniment on keyboard workstations (you press a chord -- or even one or two keys -- and the virtual band is playing the backing track in the right key).
Anyway, for your problem of having songs in your head: hum them into a portable recorder (cell phone/whatever) whenever you have an idea. At home, try to pick up the melody on an instrument that you can play. With software, record the melody to a click track/metronome. If you can figure out the chords, layer them on top. There are plenty of drum loops lying around. Boom! You've transformed an idea into something others can listen.
IIRC that's how Michael Jackson composed. And if you can't play an instrument, you can hum to a friend who can play, and figure out things together. It really works!
The constraints need some tweaking. For example, it should be possible to repeat a note by repeating a keypress. The input could be interpreted more strictly as a Parsons Code [1].
This was fun to play with, but impossible to improvise on for anything other than a relatively random sequence of notes.
While I could get a bit of a feel for how my actions would influence the piano, there is still a huge disconnect in there. Even something like trying to repeat a note can happen in three different ways (same key, key above, or key below) and there is no indication of which way will work this time.
So improvisation is limited to
- play high or play low
- change tempo
- learn what the machine will do in a situation and try to preempt it; "previously when I pressed these buttons in this order it did a thing, maybe if I do it again it will do a similar thing"
Improvisation isn't just randomly hitting notes, it's connecting different pieces together in new ways. If I play some random notes and I like them, I want to play them again.
Right now there is no way to craft music with this, just different ways to poke the machine to see what kind of music gift it gives you. I want to craft music with this, but I've no idea what that bridge would look like.
This suffers from the limitation of many generative music algorithms/tools in that you're locked into a key (in this case C major - no black notes are played at all), which makes it a cool toy and research project, but of no real use to anyone.
Most of what makes good music good (or interesting music interesting if you'd prefer) comes from harmonic interplay and shift. Bach was a master of this. If you listen to the unaccompanied 'cello suites[1] then what makes the music so fascinating is that even though only one note is being played at a time, the harmonic landscape is shifting and changing (often in 'unexpected' ways) all the time.
Doing this with algorithms and making it sound good is hard, and nobody's (afaik) done it well yet.
It's a fun idea but it really threw me off when playing around with it.
It feels like the proverb 'too many cooks spoil the broth', i.e. both me and the NN want to decide where the song is going and in the end no one is happy with the result.
I found it especially confusing when trying to play the same note twice and it changed the note, or when changing the note and it played the same note again.
But as I said it's an interesting experiment none the less!
Olafur Arnold is a Neo-Classical Musican from Ice Land. He took this idea (Minus the garage band) and made two pianos connected with a midi controller which he hired a programmer to make the software for him.
I seriously think that more and more musicians will employee programmers for musical reasons.
Here is his Tiny Desk Concert. Pretty amazing stuff.
I'm not grasping the significance of this. To my ear, it sounds like the pianos are both playing basically an echo of the chord that he's sending from the controller keyboard. Doesn't seem that much different (though I will concede, much more aesthetically pleasing) from the auto-accompany features that have been available on keyboards since at least the 1990s.
The MIDI interfaces on the pianos aren't at all novel, either: any piano can be retrofitted with a PianoDisc system. Just about every pipe organ nowadays also has MIDI capabilities.
Well if you watch him play he isn't playing a simple chord that is controlling two pianos. He actually has them randomly produce different sequences periodically. He plays a chord and the two pianos have two different arpeggiations going. If you look a little closer you will see the sequences coming and going.
> The Icelandic multi-instrumentalist and composer's fourth solo LP finds him working with Stratus, a piece of computer software he programmed over two years in collaboration with Halldór Eldjárn. It uses MIDI values triggered by what Arnalds plays on his primary piano (equipped with a Moog Piano Bar) to generate different sequences on two player pianos. The process can create randomized feedback and lead to progressions that might be unplayable by conventional methods. Arnalds has described it as a way to change up his creative thinking: "You get ideas that you would never get otherwise, just because this inspired those ideas." https://www.residentadvisor.net/news/41922
> The MIDI interfaces on the pianos aren't at all novel, either: any piano can be retrofitted with a PianoDisc system. Just about every pipe organ nowadays also has MIDI capabilities.
Well if you listen the pianos they have a totally different sound and it is due to the mechanical way they were designed. He has stated that he had spent tens of thousands of dollars and countless hours to get the two pianos to sound this particular way. This isn't some simple little trick that anyone could pull off.
This is really neat. I like that they mention Rock Band and Guitar Hero as those are the first things I immediately thought of for this.
Serious question, though, that I hope will start some debate:
I know some people that would consider this to be a hinderance to music but I, personally, think it's great that this could potentially be used to bring people into music that normally would be too intimidated to consider it. Is this still considered "playing music" since the person using it is still making choices and decisions even if those choices are aided by the computer algorithm?
At this point, there are so many ways that a person can go about making music that I (as a person who makes money every week playing music) don't feel comfortable saying what "counts" as "playing music"... if people enjoy it and it's musical, then it's a win.
Of course... I agree. I just constantly see people say things online like "DJ's aren't real musicians" and "people who play those sample pads aren't really making the music, they're just picking the order it's played in" or things like that. Just curious as to whether people who think that have an actual argument behind it.
> Is this still considered "playing music" since the person using it is still making choices and decisions even if those choices are aided by the computer algorithm?
Think about hang drums, diatonic harmonicas, or about the Tenori-on. They make it basically impossible to play a "wrong" note.
But a normal piano was already a step in this direction.
A piano restricts the pitches to discrete ones, making it harder to play a "wrong" pitch.
Violins, and to some extent, guitars, have continuous pitches. And there are indeeded some pitch nuances that piano players can at best hope to hear (e.g C# vs Db), and some pitch effects that a piano player can only mimic (glissando, vibrato).
However, nobody today would argue that a piano player isn't really "playing music".
> this could potentially be used to bring people into music that normally would be too intimidated to consider it
And this is where the real issue lies IMHO: playing or making music shouldn't be "intimidating", but the way we teach it certainly doesn't help.
Pianos evolved from older instruments where each string had fixed pitch - probably psalteries, which are related to harps - so there was no stepping towards removing wrong notes. They were simply different instruments. In fact the stepping went in the other direction, from a limited note selection in one scale to a full chromatic scale that could play in any key, wrong notes included.
Many things that humans do are intimidating - such as running a marathon, climbing Everest, or inventing Smalltalk.
Removing challenges devalues the personal satisfaction of mastery. Making any art simple makes it trivial and unexpressive.
So you can't say anything of interest or value with machine music that does most of the musicking for you. You can make some sort-of-music-like noises, but that's all.
> Pianos evolved from older instruments where each string had fixed pitch
Interesting. But my point is actually unrelated to history.
It's related to the tradeoff between "easy to play" and "expressive power":
There are some steps you can take, as an instrumentist, in the "easy" direction, while still keeping enough expressive power.
> Removing challenges devalues the personal satisfaction of mastery.
Surely. It's unclear, though, how instrumentist mastery correlates with musicality:
Some virtuoso guitar solos can be considered "non-musical". To the other extreme, some great composers don't use any musical instrument (from classical music composers, to Amiga sequenced-music composers) - some of them can't even play their own compositions properly!
> Making any art simple makes it trivial and unexpressive.
This reminds me of an old a running joke on french TV, starring a fake David Guetta having a custom piano built, with only one key ( a reference to the intro of "I Gotta Feeling" ). I think it's safe to say that you can't express a lot with a such a piano :-)
However, a smart 10-key "piano" opens new horizons: because now it's reasonnably possible to "play" multiple pianos at the same time.
I can certainly see new forms of mastery and non-trivial expressiveness here (e.g a setup of 4 "smart keyboards", each of them using a different sound).
This doesn't invalidate the point you're making, but saying that it's 'basically impossible to play a wrong note on a diatonic harmonica' is simply false, as anyone hearing me play one will attest to.
I think this might work a bit better if it were trained on a bunch of popular songs from a large fake book and tried to guess the one you're trying to play?
As it is, you're never going to get a melody or accompaniment out of it.
And there's plenty of data available - Project Midomi[1], for example, has accumulated enough data to recognize popular music by (often out-of-tune) humming.
The next step would be doing it in real time, latching onto your humming and providing a backing track!
I sounded like a drunk Jazz musician. (Not that drunk Jazz players are necessarily a bad thing.)
Is there an ABBA mode?
I actually once noticed that melodies of popular tunes tend to hover around the chord keys. The duration and probability of a given note favors chord keys. I built a Compose-A-Tron based on this pattern, and it mostly worked, but probably wouldn't generate any #1 hits. (Although, it could be improved with tweaking; but so could ordinary draft "on paper" compositions.)
I think part of my disatisfaction is due to pattern recognition. As humans we recognize patterns EVERYWHERE, even when they don't exist. So what happens (or what I believe might be happening) is that I think I recognize a pattern, try to play a note to continue it, then the NN decides otherwise. Although the resulting sound may be nice to an outsider it breaks my internal model, making it sound "bad" to me.
I do however like the idea of augmenting insturments to make them easier to play, I'm personally guitly of having many potential songs in my head but not having near enough the skill to execute them.