Hit Song Formula.

[MikeRobinson]

The key thing to realize about "AI" is that there is still "a human being(!)" who is very much involved in what comes out of it.  "AI" is still based on a computer algorithm, known somewhat-euphemistically as a "neural net."  It is basically "a network of weights," based on our growing understanding of "neurons."  These algorithms are not being expressed in terms of "if-then-else," but they are nonetheless "computer algorithms."

This algorithm is still very-much being programmed, now through an indirect process called "training."  Where the intended goal is definite, a neural net can navigate through quite an interesting "mine-field of uncertainties" to (sometimes ...) get there.  For instance, your phone has become good at recognizing your voice, and at weeding out your "spelin erurs" in both a word-wise and contextual manner.  "So far, so good."

"Run-of-the-mill 'pop songs'" might well have become "a definite goal."  But, "songs in general," not so much.  After all, if it were actually that easy, our human brains would have achieved this long before now and we'd all be rich (or, poor).

Nope: we've not yet achieved what was in those old Peanuts cartoons:  "Did you know that Beethoven comes  in a spray can?"

Edited January 19 by MikeRobinson

[des99]

33 minutes ago, MikeRobinson said:

"AI" is still based on a computer algorithm, known somewhat-euphemistically as a "neural net."  It is basically "a network of weights," based on our growing understanding of "neurons."

Let me just jump in before we propagate some misunderstandings there.

The majority of these "AI" tools in practical use today (at least the ones in the context of this thread) are based on machine-learning, using training data to build a model that can perform work. This is the bulk of the AI we're talking about in terms of ChatGPT, Stable Diffusion, AI-generated art, music, language, computer programs, and all that other stuff. (I'm keeping things simple here, there is other stuff at work too.)

This has nothing to do with neurons, or how the brain works.

Neural networks are one specialised type of AI method, that is designed around tasks where you want to process data or perform some task similarly to how the human brain works. And while the tech has come on quite some way since I studied AI at university, neural networks are not the basis of these machine learning models we're talking about.

The suggestion above is that AI = neural networks, all based on modelling neurons, which is just not the case. Neural networks are one small subset of a whole range of technologies in the AI space, the most promising of which in modern times is machine learning, and that has nothing to do with neurons.

For more info for those interested, just google something along the lines of "machine learning vs neural networks".

As a shorthand, neural nets can make decisions and inferences on their own, whereas machine learning can only make decisions based on learned training data. It's more involved than that of course, they are quite different techniques, but it's a useful simplification.

Edited January 19 by des99

[MikeRobinson]

Thank you for this clarification, @des99 ... yes, it is an important one. I guess I was trying too hard to "simplify," and in so doing left out quite a bit.

The key term in this case might be: "learned training data."  An entirely different sort of computer programming, but "programming" nonetheless.  This is non-procedural computer programming:  it is no longer just a matter of "if-then-else," but it is still programming even when the algorithm seeks to "program itself."  Where the problem domain can be well-defined, the outcomes are very exciting indeed: results far better than "procedural" method could hope to come up with, with [potentially] much greater reliability and much(!) less cost.

For a "pop music industry," this might initially look like "a gift from heaven" – until it doesn't.  So it goes.  (Now, me and my "girl" in her "tight blue jeans" are gonna "go down by the river" in my "red pickup truck," either "until the sun comes up" or "when we've got to sneak past Daddy on the front porch with his gun."  I'm sure that your "algorithm" will understand ... why I just changed the channel.)

Edited January 20 by MikeRobinson

[des99]

Indeed, your description of "weighting" is more or less how it works, for ML-type applications (though there is a lot of complexity in the details, of course).

The regular computer programming - the if-then-else style, is of course the computer code used to build the system, the same as any regular computer app. But then it uses maths and lots of data, and usually some kind of supervised learning, to build a model from a set of training data, and that model can then be applied and used.

Obviously, how the system is designed and works, and the algorithms used, influence the outcomes, as does of course the training data used in various ways and algorithms - I see Getty today have launched the first of the (inevitable) lawsuits against Stable Diffusion, Midjourney et al, for unlawfully scraping and using Getty Images in their training (it was a bit of a giveaway that people were finding AI generated art with synthetic Getty Images logos on them... oops!)

Edited January 20 by des99

[gacki]

I just stumbled across this one:

 

It's really fun to watch (for me at least - I laughed out loud when I saw ChatGPT struggling with the concept of a key for a given progression). It's also interesting to see how quickly ChatGPT threw in the towel when being confronted about the chord progressions not being correct for the given task.

[Little Fat Bloke]

Are things like this any good for 'thinking outside the box' when it comes to chord structures?

I'm becoming more and more frustrated with my lack of adventure when putting a chord sequence together. If I write a song in G then C - D - Em - Bm etc will inevitably make an appearance, and when I'm only using guitars, bass and drum patterns, things quickly begin to start sounding stale and predictable.

[des99]

1 minute ago, Little Fat Bloke said:

I'm becoming more and more frustrated with my lack of adventure when putting a chord sequence together. If I write a song in G then C - D - Em - Bm etc will inevitably make an appearance, and when I'm only using guitars, bass and drum patterns, things quickly begin to start sounding stale and predictable.

Chord hack - to extend your harmony options without a huge amount of extra effort and theory, think using regular triads, but with a different root note. And explore some 2nds, 7ths, 9ths, and 11ths for colour. Also work in different keys that you are less comfortable playing, so you don't keep resorting to the chords your brain already knows kinda work.

By far the simplest, quickest way of developing chord structures is playing them directly on an instrument, because you can just immediately try stuff by moving your fingers around, rather than keep flipping back to tools to generate new options, and then sift through them looking for something that might work.

I get that perhaps for people with zero instrument or music theory skills, they just want the computer to give them something they can use, but it would drive me nuts to work that way...(!)

[Little Fat Bloke]

22 hours ago, des99 said:

I get that perhaps for people with zero instrument or music theory skills, they just want the computer to give them something they can use, but it would drive me nuts to work that way...(!)

That would take the whole creative process out of songwriting, getting a machine to do it for you.  I couldn't work that way, either. Press a button and you've got a song...like a commodity being picked from a conveyor belt. Like a 21st century Stock, Aitken and Waterman.

[gacki]

6 hours ago, Little Fat Bloke said:

Like a 21st century Stock, Aitken and Waterman.

S, A and W and in Germany Modern Talking and the other Dieter Bohlen acts really got on my nerves back then. Talk about formulas. What really irks me is how S, A and W sometimes tried to connect/justify their assembly line technique to/with Motown. Yes, Motown (and other labels at that time) also did stuff that was to a degree pretty formulaic - but they always tried to extend on that and make it sound "different". Do I recognize a certain drum fill in numerous Motown productions? Of course I do - but it's a small part of an ever varying picture. And that's what I found sorely missing from the alleged disciples.

Now, KLF's "The manual" in a certain way showed how music production could be done if your desire is to be famous but without little ideas of your own: ask other people to supply those ideas and then decide which one is best. Making those decisions then makes you the songwriter, right? I've once participated in such a session myself and vowed afterwards I'd never do it again. I could imagine that AI generated music might appeal to that particular demographic.

[Nunstummy]

7 hours ago, Little Fat Bloke said:

couldn't work that way, either. Press a button and you've got a song...like a commodity being picked from a conveyor belt.

That’s what coming, like it or not.  Production music is getting hit first, putting composers for tv and commercials out of business - according to friends I have who do that.  Instead of hiring a composer for a 30 second ad, there are AI apps that generate music based on a theme an genre.  If you’re the production company and you can generate original music for $50, instead of paying a composer $2000, you’re going to consider the lowest cost option.  Plus, tv viewers don’t care about the music on ‘The Goldbergs’ or an ad for Downy Laundry Freshener.

[Little Fat Bloke]

Not just manufactured music, but manufactured artists as well. Going back to the 50's (in the U.K. at least) with people like Larry Parnes 'inventing' Billy Fury, Tommy Steele etc. Even up to the evil overlord himself, Simon Cowell and his stable of disposable pop stars.

It is extremely depressing to think that some computer geek somewhere on this planet could well be creating an app that will eventually replace writers, arrangers/producers, musicians and so on all in the name of profit.

 

 

 

[gacki]

On 1/25/2023 at 12:05 PM, Little Fat Bloke said:

Are things like this any good for 'thinking outside the box' when it comes to chord structures?

I doubt it. Gaining additional knowledge, using permutations and substitutions and so on. I'll give a real-life example further down. Case in point:

On 1/25/2023 at 12:05 PM, Little Fat Bloke said:

If I write a song in G then C - D - Em - Bm etc will inevitably make an appearance

Nothing wrong with this; but here's a couple of ideas. The whole thing strikes me more Em-ish than G-ish so I'll work under that assumption for now. The whole progression has a nice "loop around" feel which would make it easy to jam over it; I want to preserve that, too.

So Em is the focal point here; the tonic, the I, the tonal center - whatever you might want to call it. So it stays for now. Can we reach it in a different way? Of course we can. Let's swap the D with the Bm and so we get

C - Bm - Em - D

This would very likely still work with most melodies but gives the whole thing a different feel. The jump upward to the tonic emphasizes it a bit more. I'm not completely happy with the steady downward motion leading away from the Em but let's leave this for a moment.

Why not use a full fledged major dominant?

C - B - Em - D

This gives stronger voice leading (the D# in the B chord leading to the Em). So why not? Or let's go back to the original and substitute the B for the Bm there:

C - D - Em - B

This gives a chromatic movement of the B chord to the C chord (and would be called a "deceptive cadence" or "Trugschluss" in the key of Em).

Let's mess around with the first chord instead.

Am - D - Em - Bm

This gives stronger general movement because the roots are further apart. If desired this would also gives us the ability to go Bm - Bbm back to Am if so desired (but this depends a bit on the type of music we're dealing with).

I've limited myself to very basic substitutions so far. In real life I'm a complete sucker for non-root bass notes and pedal points so here's one version that works very well for me and combines a couple of ideas:

C - B/D# - Em - Bm

Strong resolution through the bass leading from D# to E, and afterwards a nice fall to the Bm and then a restart of the whole thing.

And yes, that's the stuff arrangers are trained to do: Look at the material and check what else can be done with it. I already mentioned Coldplay's "Paradise" for an example of true and very original reharmonization; but let's have a look at another song: Ellie Goulding's "Burn".

The whole verse (and chorus) go like

Bbm - Gb - Db/F - Ab

And the prechorus starts off exactly the same. The melody line is sung identically three times - but during the third time there's a subtle shift in the harmonies:

Bbm - Ebm - Db - Ab

which leads into the final line of the prechorus with the chords:

Db/F - Gb - Bbm - Bbm

(Let me say that this song is a bit of a "guilty pleasure": I find the vocal effects somewhat obnoxious; and if I had been asked I probably would not have ended the prechorus on the same chord on which the chorus starts - this struck me always a bit as a cop out. But if I had been asked it probably wouldn't have become a hit, so yeah...).

This slight reharmonization in the prechorus immediately reminded me of my arrangement class: If you repeat the same thing three times in a row do something different with it the third time.

(Also interesting: after the last chorus the prechorus is sung again - but this time completely over the main chord progression. The melody still fits perfectly.)

Adele's "Rolling in the deep" does something similar.

Those are little things that certainly don't make any of those songs "harmonically complex". But they make everything a tiny bit less predictable and uniform. So when I'm doing arrangements this is something I'm using often.

[MikeRobinson]

Now, if you want to encounter an honest-to-god textbook, buy a copy of Jimmy Webb's (yes, that "Jimmy Webb ...") "Tunesmith."  My two copies, each autographed(!) by the Master himself, are among my most-prized and most-thumbed possessions.  I re-read them at least twice a year.

One of Professor Webb's key insights is that, when each chord "moves" to the next, three or four individual notes actually "move" together.  Each of these movements, in relation to whatever chord happens next, is musically described as an interval.  And, even though your untrained ear doesn't necessarily "know" what it is hearing, "it hears it, nonetheless."  It hears both "tension" and "release."  One or more particular notes "carry" the transition from one chord to the next, but you can actually select(!) which note(s) that will be.  There are actually several "chord substitutions" which could work.  And, as soon as you understand the "trick," you can begin to experiment.

(Obviously, "a keyboard" is hugely beneficial.  Even if you don't "usually" play one.)

I certainly agree that "Coldplay" is one of the best current practitioners of this art. (Who can answer "Clocks?") But I still continue to be impressed by examples like Leann Rime's "Can't Fight the Moonlight."  (A song written for the entirely-forgettable film, Coyote Ugly.)  In which different harmonization was applied around a basically-unchanging vocal part. Although they use a few "overdubs" in the final pass, her singing really doesn't change. But the feeling of each key phrase certainly does.

I really don't fear AI at all.  Because, "Beethoven now comes in spray cans."

Edited January 27 by MikeRobinson

[Little Fat Bloke]

13 hours ago, gacki said:

C - B/D# - Em - Bm

Now this is something I can relate to because I use it so often. The B/D# obviously takes us to the E note.

Effectively, the chord sequence of a song is its foundation, alongside the melody: and what I love about it is the journey those chords take us on. From the first chord to the last it's like exploring new frontiers, changing root notes etc. Artistically, at the end you can say to yourself: "Wow! That was some journey!"

[MikeRobinson]

P.S.:  I guess that when I was describing "Professor Webb's insights," I actually failed to complete my sentence ... and I actually got it wrong.  What I meant to say (note the "strike-outs" and [omissions]  ...):

"... when each chord "moves" to the next, [three or four] individual notes actually "move" together, while one or more of them do not."

The notes that do not move provide the "anchor" against which the others are now perceived to move.  And, if those same notes continue to not move in subsequent chord progressions, the "tension" and "diversity" is naturally increased – both against "their predecessors" and against "the anchor."  This will continue until at least one of the anchor-notes "finally gives way."

Therefore, in any given chord progression, or sequence of chord progressions, there is both "tension" and(!) "release."  All of which might extend across multiple measures.

Edited January 31 by MikeRobinson

[Nunstummy]

20 hours ago, MikeRobinson said:

the "tension" and "diversity" is naturally increased

Good technique.  I used that in a song I composed called "Un-Resolved", which the singer interpreted as a relationship that ended without being resolved.  It was renamed "Above All Else" for the album release.  There's no last chord in the song on purpose.

The song takes 13 seconds before it starts.  At the time, we thought this was clever, but nobody waits 13 seconds for a song to start.  CD sales were strong in 2007, but streaming revenue is nothing.  After the project, there was no promotion of the work other than live concerts.  Old-school.

[gacki]

On 1/31/2023 at 9:52 PM, MikeRobinson said:

"... when each chord "moves" to the next, [three or four] individual notes actually "move" together, while one or more of them do not."

The notes that do not move provide the "anchor" against which the others are now perceived to move.

Maybe there is some additional meaning (or background) lost - but I can think of tons of chord movements without any common notes - in other words, where a chord is simply moved up or down one or two half-steps. Subdominant and dominant triads don't share any notes.

[Little Fat Bloke]

This topic has proved to be quite an eye-opener for me, in terms of theory and so on. I don't read music so a lot of goes over my head, admittedly, but it's still useful to pick up a few tips and tricks from you musical scholars!

[MikeRobinson]

For certain genres of popular music, it's still true what Lucy once said to Schroeder: "Did you know that Beethoven now comes in spray cans?"  😀

A certain segment of the industry simply needs a large and reliable supply of "product."

There's always going to be steady work in Nashville for someone who can play a steel guitar, and for songwriters who can quickly craft another "three chords and the truth" melody and a squeaky-clean lyric about a woman – inevitably named "girl" – who wants to ride in a horny boy's "[white|red] pickup truck" "down by the river" to do what is never directly mentioned.  If you can reliably appeal to the slightly(?)-intoxicated young women who can be found wearing leather boots and cowboy hats down on Broadway or in The Gulch on any Saturday night in "Nash-vegas," you will remain gainfully employed.

But, even within that tired genre, there is the occasional Tunesmith ...

Edited February 7 by MikeRobinson

[gacki]

Interesting:

https://www.abc.net.au/news/2023-02-15/google-ai-music-quiz-musiclm-humans-artificial-intelligence/101967746

I won't spoil the fun to give everyone interested a shot to try it out first.

I got 8 out of 10.

[FLH3]

I got 7 on 10.

[grumblepig]

8/10, but the scary thing is that the examples presented ALL have a generic quality that rubs me so very the wrong way, grumble grumble. 

[MikeRobinson]

The only "formula" is: "there is no 'formula!'"

[Little Fat Bloke]

I have a formula for my songs: bland, tuneless, dreary and clichèd. It's a talent I have.

[FLH3]

I've another formula for my greatest success, as powerfull as yours, I think: Strings, reverb, more strings and more reverb, with strings and reverb on the top.

And sad, if possible. 😄

[Nunstummy]

5 hours ago, MikeRobinson said:

The only "formula" is: "there is no 'formula!'"

Cliche - although NOW we’ve got AI programs that promise to deliver the very essence of this discussion.

[grumblepig]

Just now, Nunstummy said:

Cliche - although NOW we’ve got AI programs that promise to deliver the very essence of this discussion.

I've been playing a lot with AI in recent weeks. I asked a chatbot program to suggest some uncommon chord progressions. It offered up what I'd say are, in fact, some of the MOST common progressions... I wrote back suggesting such. It apologized - and agreed - then offered far more interesting progressions on its second attempt. Not to say these new progressions were radical or weird, but they were, for me, not what my hands would normally reach for, and certainly not as familiar as the sort of Ed Sheeran AirB&B tones we hear daily...

[Nunstummy]

Does it make any difference?  I can’t help but think of the Flintstones episode when Fred and Barney decide to write a hit song.

but the wisdom of Hoagy Carmichael saves the day…

[Nunstummy]

3 minutes ago, grumblepig said:

I asked a chatbot program to suggest some uncommon chord progressions.

If the goal is to craft a popular song, wouldn’t you want the MOST common and well liked chord progression?

Let’s face it.  We might not like what’s popular.  I know I don’t.

[gacki]

1 minute ago, Nunstummy said:

If the goal is to craft a popular song, wouldn’t you want the MOST common and well liked chord progression?

Ideally you'd want something that is both familiar and also stands out; read: incorporates something uncommon.