-2

Below I document my thought process, which is obviously wrong, else one of the billions of people would have commercialised it.

Assumptions:

  1. A good piano player (let's substitute piano for instrumental hopefully without loss of generality) can get a note-book as input and produce a mostly accurate audio rendering of it (output).
  2. The major product that a piano player manufactures is an audio rendering of the note-book. Show and autographs come second.

Observations:

  1. For the late 20 years, there have existed multiple computer programs that perform this. They even allow the user to chose the type of piano.

Limitation of scope:

  1. This question excludes live performances as there the artist dancing is a major factor. It considers only studio recordings.
  2. It also excludes composers and other higher order artists. I believe playing an instrument is the way to learn to compose for it.

Questions:

  1. Why are instrumental performances still a thing? Why spend 10 years to be good at piano, when a computer can spend 0 and be perfect?

Antithesis1: The beauty of music is in the small mistakes an artist makes. My response: noise can and has been statistically modeled since forever.

Antithesis2: The years of effort the artist has put into learning the instrument makes the music all that more meaningful. My response: IF we only care about the output audio file or acoustic vibrations, performance is not important.

This question has plagued me for years but not being an artist myself there's zero chance to answer.

6
  • Comments are not for extended discussion; this conversation has been moved to chat.
    – Doktor Mayhem
    Nov 22, 2020 at 22:56
  • 3
    Either a duplicate or a very related question: music.stackexchange.com/questions/67803/…
    – Dekkadeci
    Nov 23, 2020 at 12:19
  • 4
    What if what you call 'small mistakes' are signal rather than noise?
    – AakashM
    Nov 23, 2020 at 15:43
  • 2
    @AakashM, this is an excellent point. I think the OP is confusing many topics and there are some false assumptions here.
    – user50691
    Nov 24, 2020 at 23:07
  • I suggest to start playing an instrument - even a cheapish plastic recorder may provide tremendous amount of insight after a few hours. For the other end of the scale (the mechanical one) one get an idea by playing some MIDI files of simple children songs.
    – guidot
    Nov 25, 2020 at 11:02

7 Answers 7

8

A few points:

  • The charm of music or any art lies in the artists expressing something from their inner cognitive universe through their instruments, in an aesthetically attractive way.

  • Playing a melody perfectly by a computer is about as interesting as watching an automatic machine in a factory churn out identical items one after the other all day long -- it may be fascinating for a few seconds, but then it gets old very quickly.

  • Computers are also musical instruments. Highly powerful ones in fact! But to use them to produce real, powerful art, also requires great artistic talent and a lot of technical skill, albeit in the form of programming rather than in the form of bowing or picking.

  • Most computer-produced music is trivial lazy stuff. Putting a few notes in a row and having a computer play them at fixed intervals doesn't mean anything, see above. But I'm sure that we'll gradually see more and more true artists emerge, who will be able to produce highly innovative and amazing art with their computers. But it won't be any easier for them, than it is for a violinist to go through grueling years of training to product the final result.

  • In a sense, you get what you pay for. If one person spends 15 minutes to write a few notes to a computer and play them, while another spends 15 years exploring all the possibilities, studying, researching, learning, testing, producing... then the results will also reflect that.

  • Bottom line, for me at least, the computer is just another instrument. The triangle is a simple instrument, but even after we invented more complicated ones, we still like it and use it. Computers won't replace the existing instruments, they will add to them. And like with all instruments, half-assed efforts will produce half-assed results, while talented and hard-working artists will create new wonders with them.

2
  • I like all of these, but I particularly like the 'half-assed in, half-assed out' analogy ;)))
    – Tetsujin
    Nov 22, 2020 at 16:02
  • 3
    I think one very important bullet point to add is that computer instruments still can't make the same sounds that real instruments played by real musicians do. Even the best string section sample libraries still don't get the bowing, legato phrasing, and other details quite right. Nov 22, 2020 at 19:26
8

"Why" questions don't work very well on SE, because the answer is often 'no-one knows', however…

In this case, there is little commercial drive for it.

A musician wouldn't want it because they'd be out of a job. A songwriter wouldn't want it, likewise. The consumer wouldn't care how it was made, but unless someone was going to put a whole set of fake performances behind it to fill up the YouTube etc side of the business, who would even know it existed, let alone buy it?

That's before we even get to the 'how good can it be made to sound?' issue.
Some performances - the modern pop record for example - are tuned & aligned to the nth degree. Enough to satisfy the 8 to 13-year-old the product is intended for.
They are already swayed more by the artist's video & twitter feed than they are by the music.
By the time you get to an older, more discerning audience, prejudices have already begun to form. Trying to pass off some computer-generated algorithm is likely to offend as much as amuse.

Software already exists to generate chord progressions & even melodies, but even people using these would tend to heavily edit using their ears & learned musical sensibilities. there is also software than allows the user to build from pre-made building blocks, like Lego. This also tends to really be used by the followers, not the leaders in any field or genre.*

So, there's just no real avenue for it to develop in… as yet.

I have a friend who makes lift musak … ermm … production/library music for a living (& he makes a ruddy fortune doing so). He can turn out an album in a fortnight, so long as he doesn't have to take an orchestra into Abbey Road for the 'posh stuff'. That's probably quicker than it would be to sift through the random musings of a computer set to work on the same 20 tracks.

A real-world analogy to this. I was once told by Pete Waterman (yup, him) that if you're making a record that sounds like one in the charts, you're already three months too late. I'm not so sure that these days that is quite so true as it was in the early 80s, but there must still be some truth to it.

Late edit*
There seems to be some uncertainty as to whether the question is actually about whether a machine can compose & arrange a piece or the far simpler can it play a piano score.
If that's the case then the answer becomes, very simply…

They already can. It sounds like pants. No-one would ever use one.

It's like comparing the Duracell Bunny to a drummer.
Drum machines have their place. That rigidity was an element in early 80s music, however it was very often tempered by humans playing live over the top of it, softening the hard lines. Modern dance tracks, pop, EDM etc tends to re-use samples of things that have actually been moulded slightly differently. The origins of those samples have been themselves 'softened' from their drum machine ancestors. Slight temporal variations - whilst in themselves repetitive over time - have been shifted slightly out of true, to make them 'feel' better.

Once you step up from the humble drum machine to emulating how a pianist would handle a full piece… you are several generations of computer AI away from it being really possible.
Computers don't 'feel'. Early attempts at randomisation algorithms to emulate human inaccuracy turned out to just sound 'bad' not 'human'. They've largely been left by the wayside since.

6
  • 3
    I think the OP is not talking about computer-generated music, as in, computer-generated sheet music. He's talking about computerized playback of existing music, sort of like a player piano playing MIDI files. And the OP sees the act of playing an instrument from sheet music as trying to get a perfect score in some kind of "press the buttons in the right order at the right time" test. In other words, interpretation is an unknown concept. Nov 22, 2020 at 16:42
  • 1
    Ugh. If they are, then we can trim this entirely down to "because it sounds like pants" or "What's the point in getting anything or anybody to play from sheet music of an existing performance… when there's already an existing performance?" If that sheet music were a classical piece, who on earth needs a new version which is bound to be lifeless?
    – Tetsujin
    Nov 22, 2020 at 16:45
  • 1
    I'm less sure that a songwriter would necessarily be out of a job. Combine with an arranger/producer who generates all the parts and, if it works, it could be good for both. Could be the same person. But otherwise +1 Nov 22, 2020 at 16:52
  • @piiperiReinstateMonica indeed. Aslo MIDI is an important keyword that I should have mentioned in my question.
    – Vorac
    Nov 22, 2020 at 16:54
  • Ripostes: 1) Just because the producer is also a musician/composer/arranger does not negate the argument that the computer cannot yet replace any of them. Paul Simon, The Beatles or Bob Dylan may have replaced the traditional separate job of composer, but at the same time they replaced it with another… themself. 2) Midi is a recording/transmission medium. It doesn't in any way contribute to the sound or feel of the final track or arrangement.
    – Tetsujin
    Nov 22, 2020 at 17:06
4

Oh, this is ALL opinion and not what we like here. But...

Glenn Gould was a master of classical piano, and he recorded Bach's Goldberg Variations twice, around twenty years apart. They are substantially different, as the player was different. Some prefer the first, and others prefer the second. Both are considered "accurate audio renderings" of the compositions.

I'm not a classical musician. I'm a rock guitarist, who is in the class of musicians least likely to accurately reproduce a performance from a lead sheet. But given a harmonic and rhythmic context, I'll go for it and produce some melodic improvisation that works for it. I might not get to close the next time, or I might reuse licks that I'm sick of myself, but it'll be different every time. And while I have played to jam tracks, where everything is the same but me, playing with other people is more fun. I consider it a conversation.

"This question excludes live performances as there the artist dancing is a major factor." My oh my, that missed the point, in two ways. The first is mentioned above, and the conversation is not just between the musicians, but the audience. The live output of the Grateful Dead shows how the interplay between the audience and musicians made each performance different, which is part of what attracted people to the shows.

The other is that shows where the artist dances are the ones most likely to be as generated as you suggest: the whole thing is timed off click tracks so lighting cues, backing tracks, effects changes and vocal dubs (because singing on key when involved in intense cardio exercise can be hard).

All of which avoids your point, which is that, for any given instrument and composition, we can compile a "perfect" "performance" with an electronic model of that instrument.

But can we? The first instrument we modeled to replace performers is the drum kit. The "problem" was inconsistent rhythm; the drummer would tend to speed up during a song. A drummer can also play louder and quieter, depending on the desired feel, and, per That Thing You Do, change the feel more than any other instrument. We still use drum machine, but we still use drummers.

3

It is simply not true that "The beauty of music is in the small mistakes an artist makes." This is because there are cases where (1) the artist does not make a mistake and (2) people enjoy the non mistake parts more than the mistakes when mistakes are made. Your assumption (perhaps an unspoken implicit assumption) is simply incorrect. You'd be wise to send out a questionnaire to 20,000 randomly selected people and see what that says before assuming that this is true.

"My response: noise can and has been statistically modeled since forever." This is also false as "noise", especially white noise, is more or less static and that does not model a "mistake" in the typical sense.

There are a few reasons why computer generated performances are not currently the norm, though they may start to be in the near future.

The first is that what constitutes expression in music is still not entirely quantified. People tend to focus on the discreteness of tones in western music and cite that as being so small as to not offer much possibility for expression. However I don't know a single musician who would fall for this argument. Time, and timing, as far as we know is a continuum. That is to say where I choose to place a note in the flow of musical ideas is what really allows an infinite degree of expressiveness in music. We call it phrasing. And the phrasing is NOT random. We choose how we phrase things in time. Now, in Western music scores are discrete in note choice and in the division of a beat (usually a fraction 1/2^n). So it would appear that timing is quantized too. But this is really a fault in the Western way of thinking and even Western musicians know this. We have grown accustomed to our conventions for a couple reasons: (1) because it is a sufficient approximation of where to place notes in time and we can inject our own inflection as artists, and (2) to accommodate managing large orchestras. Orchestras need to operate like soldiers marching in step. You can't have 100 people all dropping notes behind the beat all differently. That would be a mess. No amount to "noise" in timing will mimic a masterful phrasing choice. What needs to happen is you need to understand the decision making process for phrasing.

The next reason is that not all Western instruments are frequency quantized like the piano. Many allow for bending and other slight variations in pitch and again, phrasing comes into play here. Everything done on computers is quantized, digital, and that necessarily limits the quality of what is produced.

The third issue, and this really is something we can overcome, is the timbre or variety of tone that is present in specific musical instruments and the attack of the musician. A cheap synthesizer is a perfect example of what not to do when trying to generate a fake guitar or flute sound. They all sound like synthesizers. But a very high quality synth with well modeled attack, sustain, and decay curves across a broad bandwidth of the harmonic spectrum will do a very good job of mimicking other instruments. I have worked with such high quality synths and they are impressive. To that extent live performances by orchestras in some settings are starting to be replaced with just a couple or a few keyboard players. But this is more for musical and stage performances where the music is supportive or secondary.

I would say that understanding the human side to phrasing choices, tempo changes, dynamics, etc. Is the last frontier. While most people don't like to think of themselves as evolved organic robots, that may be all we are. But we have millions of years of programming edits, updates, and patches all contributing to our complexity. Just adding a random noise jitter to a note property value would never model human qualities. But if AI evolves enough a program could be unleashed on a database of recorded performances with the task of deciphering the meaning of phrasing and other "artistic" choices made by humans. When this happens you may really see music generated by computers or robots that is every bit as satisfying as what we produce with our blood sweat and tears.

2

In addition to the other good answers, here's one more angle.

It's argued that sure, computer-generated music gets old, but the power of AI will soon revolutionize the field like we expect it to do to every other field. Soon it will be possible to create entirely new genres, new instruments that sound believable, new performances to incredible new conducting styles, etc.

However, as computer scientist Jaron Lanier argues, "artificial intelligence" is a very misleading term. It's not comparable to true intelligence. In every field, it essentially means the manipulation and extrapolation of massive bodies of data. And that data was created by the blood, sweat, and tears of huge numbers of composers, inventors, players, conductors, and so on.

It seems unwise to throw the creators of the data that powers AI under the bus by making their jobs obsolete. The input is likely to stop, and the inventiveness does run out. The newness of AI is not magic; eventually it does exhaust what can be done with the data. This is why human intelligence allows genres to change beyond recognition (comparing rock to classical, say). Similarly, this is why e.g. Google Translate constantly buys and scrapes more corpora to keep up with language change.

2
  • Isn't that we are? We extrapolate from very general data? If anything, that makes artificial intelligence seem "better" than whatever true intelligence is. I think the more important distinction would just be that we don't know what or how to tell the computer what to do (like, how do we define a "good" song?)
    – awe lotta
    Nov 25, 2020 at 1:08
  • @awelotta Who knows exactly how our intelligence works, but it seems unlikely that it mirrors how artificial intelligence works with the data. I think it's certainly more than just data manipulation, and possibly less sophisticated than the data manipulation that AI can do. I think we do ohter things than sort things into categories, look for patterns, and grade innumerable candidates so that the best surface; and I don't think we do that with as much facility as a good neural network. Either way we end up far more inventive than current AI. Nov 25, 2020 at 1:54
2

Keeping it simple, I'd say both the assumptions are wrong.

A good piano player (let's substitute piano for instrumental hopefully without loss of generality) can get a note-book as input and produce a mostly accurate audio rendering of it (output).

A pianist's job is generally to take the written score and add a bunch of extra small-scale but creative and stylistically-aware choices to make a stylistically-suitable product. As an analogy, think of a film script or a play - can the actors produce an 'accurate rendering' just by reading out the words? Rarely.

In other words, playing exactly what's on the score won't usually produce a subjectively pleasing product.

The major product that a piano player manufactures is an audio rendering of the note-book. Show and autographs come second.

I don't think this is true - I think the feeling of connection that comes from listening to a human performer is very valuable to a lot of people. Most music fans want someone to be a fan of - it's part of the fun.

2

I disagree with the assumptions already. While the ability for an exact reproduction of the score is a requirement for a pianist, the result is unlikely to be considered as music in the sense of meaningful interpretation. More steps are needed starting from that base by the performer.

One can get an idea of the result to be expected from an "exact reproduction", when playing the output from a music notation software program. One could of course apply a slight de-regularization to get a less mechanical reproduction but is unlikely to achieve something called an interpretation, because for that all the minor deviations (as opposed to small mistakes from Antithesis 1) have to point into the same direction.

It has also to be pointed out, that the few volume modifiers (ppp to fff) in sheet music simply are insufficient for shaping a piece.

Not the answer you're looking for? Browse other questions tagged or ask your own question.