Fireside Chat: Music and AI, Part 4: Teaching AI to Write Lyrics: Deep Dive into Chamber Choir Music

In this five-part series on music and AI, filmed at the 2024 Marketing AI conference, I sit down with composer Ruby King to discuss AI, music, and the future of creativity. This is part four. Teaching AI to write lyrics, a deep dive into chamber choir music. Okay. So let's do this.

Let's transition here. I'm going to go ahead and share my screen, and we're going to see how well. Yay, Infinity Tunnel. Let's go here to uh first Suno and then AI Studio. So I'm gonna use Google's AI Studio.

And the first thing I want to do is I want to get an understanding of what constitutes good songwriting. So I'm gonna ask you to evaluate the models thing. So your topic is best practices for what genre of music do you know very well that has lyrics? Pop? Rock?

No. Um I guess it's more grunge. No, it's more it's just it's classical pieces that have lyrics. I mean Gregorian chants. No.

Uh chamber choir music or something like that. What is four parts or that kind of thing? I can do that. Lyrics to chamber choir pieces. So I'm saying that is what we're gonna go for.

Um, we're going to choose the Gemini model. We're gonna choose an output length of 16,384 tokens. Does it have to be with words for this one? Because you're looking at it. Well, we want to do stuff with lyrics.

Okay, yeah. Um and the reason we want to do it with lyrics is is specifically because for tools like Suno in particular, part of the generation is the lyrics themselves. That's part of the prompt. Okay. So let's see what Gemini has to say.

Best practices for generating lyrics for chamber choir pieces. Understanding the vocal capabilities and limitations of a chamber choir. Voice types and ranges, blending and balance, ensemble size and dynamics. Text selection text, choosing appropriate texts, considerations for length subject matter, word painting and musical expression example. Word painting, that's good.

Okay. Syllabic versus melismatic setting. Yeah. Melismatic is when there was that singing before where it had no words, and it would be like AI doesn't do that terribly well, but it's it's singing multiple notes to a sound. Okay.

Creating a unified and engaging musical structure, form and development, strophic through composed ABA, vocal lines and harmony, repetition and variation, practical considerations of performance, score, clarity and readability, accessibility and singability. Common mistakes of less experienced practitioners, writing ranges that are too wide or uncomfortable. Yep, absolutely. Neglecting vocal blended balance. Okay, overlooking dynamics on ensemble size.

Oh, yeah. Choosing texts that are unsuitable for choral music. Yeah. Overusing or misusing word painting, inconsistent or inappropriate syllabic choices, lack of musical structure development, a weak or unclear form, uninstalling or unbalanced vocal lines, neglecting practical considerations. Chamber choirs can only sing quiet introspective music that is generally believed to be true but is actually false.

Rhyming lyrics are essential for chamber music. Over there. Um I wouldn't say so. A lot of the time we're singing just words straight from the Bible. Right.

So it's say they're saying these are things that people believe to be true but are actually false. Okay, yes, that's true. Latin texts are always more appropriate. Uh yeah, okay. Right, yeah.

Things that are believed to be false but are actually true. Simple texts can inspire profound choral music, silence can be a powerful element, collaborating with the living poet can increase in enrich the creative process. Tips and tricks. Listen deeply to singers, sing the parts yourselves, study a choral scores. So these are all things you would you would say are appropriate for uh what you should know to to create good chamber music.

Yeah, I mean I've not seen anything I disagree with so anything missing that you know of that it didn't know. Um did it speak about the spaces that things are being sung in. I didn't see that there, but it does also like if you're singing in a really big church, you might choose the amount of singers might be different, or like how many parts you split into. No, because we speci specified writing um lyrics. Yes, yeah, okay.

Um yes, lyrics also matter here though, because if you've got chamber choir is four parts, usually. Um so if you've got if you're writing lyrics and you're in different spaces, you might want to write more words in a shorter period of time with lots of people doing different words if you don't care what the if you're in a church service, usually when you're hearing chamber choir, you want to hear the words that are being said because it is about worship, and that is the idea. If it's a concert of chamber choir music for a different purpose, then you might have lots of different interweaving lines. Okay. Um, but when it was for church services, they'd mostly have everyone singing the same word at the same time but different notes.

So it was nice to hear, then a few like passing notes in emphasizing words, but it it does change with the purpose of the service or the purpose of the music. Okay. So next, let's have it create a scoring rubric. So next, create a scoring rubric that can evaluate a set of chamber choir lyrics, and how well they adhere to our best practices and tips and tricks so far. The individual items in the scoring rubric should be scored as variable integers.

Zero to three, zero to five, zero to seven, etc. The rubric hide that because I can't see what I'm doing. There we go. The rubric as a whole should total 100 points. The purpose is to give lyrics an assessment of how well they conform to best practices.

So let's see what the model comes up with. So it's gonna say here is how to score a set of lyrics. Oh boy, okay. Well, this is this the uh scoring rubric is one of those magic terms in generative AI where you're saying to the model, I want you to evaluate um instead of just saying model this is good or bad because the model doesn't know you you give it a a set of scores that then it can use as a as a baseline to say here's how what I think I understand about this. So let's save this.

This is our Gemini chamber choir lyrics gem uh rubric. Okay. So we've got our rubric now. Yeah. Now let's have it generate some lyrics, some chamber choir lyrics.

What topic should we use? Uh usually it's sacred. Okay. Let's not offend anyone. Okay.

And remember that this is a generative AI model, so it can make anything about anything. Okay. Um chamber choirs are usually used for slow, solemn things. So if we do usually it's quite effective if it's like saddle crying. So it just is the it's a you can have more upbeat ones, but usually that's all right.

So let's do this. All right. Next, generate two candidate sets of lyrics. The overall topic for the song should be a cat that is prowling around the streets of London hungry. Okay.

The lyrics should reflect this general situation and tell a story. Generate two candidate sets of chamber choir lyrics for this story. London's Tomcat Prowl. Soprano One, Auto One. Did it just start the word with shh?

How you meant to sing sh. I mean, okay. Some of the time it does just write that and then you make the sound. Like we had music that we literally had to all go quack at the same time with no notes. That's just what happens, but it's not a strong start.

Okay, so we have candidate one, London's Tomcat prowl. Oh, they've got it in different part. Are they split it into okay? Soprano one and two. Okay, so split the sopranos and altos and tennis and basses into one and two.

Okay, that's okay. Okay, now score each candidate according to our rubric. I did great, thanks for asking. Oh, it's quite mean. Oh.

So the first candidate scored seventy-nine out of a hundred. Second one scored eighty-two out of 100. Say great. Make a refinement of candidate two that improves the overall score. Create the candidate refinement and score it.

So what we're doing here, this is a three-part process that we use for generative AI. When we're doing anything commercially with it with generation, which is uh create score refine. One of the things people do the most wrong with generative AI is to say, just do the thing. And and they accept the output, and then they're like, oh, it kind of sucks. Generative AI is not very creative.

Well, no, you failed to a prime the model, you failed to understand the context, and then you gave it a prompt that was stupid and insufficient. And so what we're doing here is we're basically telling the model you need to keep improving until you get it right. So my mouth teaches us to say rubbish in, rubbish out to the side. So here's candidate two refinement, hungry eyes on sharing cross. Um you can see it's revised for better sound, revised for smooth line, and so on and so forth.

Um, and so this refinement it now scores an 88. Okay. Um make it even better. Yes, essentially, say great. Take the refinement and improve it until it improve it so that it scores better.

So this is candidate set to refinement too. Revise for alliteration and imagery, revise for more specific imagery, stronger association, stronger metaphorical language. Oh, it's scoring a little better now. Mm-hmm. And it tells you the justification.

Hold on, I got eight out of seven there. That's cheating. It is cheating. So it scored a 90. Now, the next thing we want to do.

So every almost every model that's made out there, like Sunos, has at some point probably used a generative language model as its foundation. So what is what are the characteristics uh how would you sell someone else to compose a piece of chamber music? That's very difficult. Um you mean lyric specifically. No, no, the music.

Oh, okay. The music itself. What would be the musicality? I'm gonna save this um the lyrics themselves, and I'm going to uh boil them. Actually, I need the final version.

Show great. Show me the final version of the refined lyrics without any of the notes for improvements. So what we want is just the clean lyrics. Okay, so we have our final lyrics here. Um that's our song title, Hungry Eyes on Charing Cross.

How would you give instructions to a composer to write the music that goes with this? Um word painting. Okay. Uh bearing in mind what have we got beneath the girls like soundbook clean? I should they lines look about equal.

So probably write a melody that aligns with the word painting of the idea of the piece, because if it's gonna be repeated, it can't work for every single line necessarily. Um use that uh in the way it's written, it might it might just be an AAA structure here. It's hard to tell. Okay, let's ask it. Okay.

See what it s what do you know about composing the musical accompanying these lyrics? Emotive atmospheric, quiet introduction, tenors and basses, soprano, alto tenor, all voices. Is there instrumentation in chamber quirks? Because it's all human singing only. Um there can be, but chamber choirs usually don't have it because there's so many there's so many different parts and all the voices kind of work together.

It's a bit more like a string quartet. Like you can have other instruments with a string quartet, but the string quartet is of its own a string quartet that is capable of doing pieces on its own. Got it. Okay. What instructions would you give a composer to generate the music for this?

Focus on technical specifics like tempo and key. So we're gonna essentially ask the model to distill down the instructions that it would give to a composer, because when we get to the SUNO portion, we want to feed it very specific instructions that hopefully will evoke the correct tokens in its memory to pull together its knowledge. The challenge is Suno's prompting structure is intentionally very short so that you can't give it highly specific instructions to replicate copyrighted music. So it has to be distilled down. So great.

Distill down these instructions to 200 characters of text. Omit verbs. Yes. Minor key is tempo shifts, good. Word painting, yes.

Dynamics. Pianis motor vortizing. Okay. References to the content, such as London night. So that our 200 characters of instructions are devoted to the music composition.

Also, why does it want an unresolved ending? I don't know. It seems to think that's important. Okay. So that's the unresolved question.

I didn't consider cello or piano optional. No. We're not doing that here. Nice dry. Okay.

So let's copy this into seven structures. We're going to remove that because we've decided that's not going to be something we do. It's actually written out PNSM and Vortissimo this time. Yes, because we gave it more room. So this is 255 characters, so we might have to as long as Suno will Suno understands.

Yeah. So we're down to 227. Very texture. Let's get rid of the unresolved ending. And now we're at 201.

Minor key chamber choir. Isn't it structural to change acquire elsewhere? Yeah, it's right down here. So we're at 187. So now let's go into create mode.

We're going to do custom. We have. Oh it's sorry, it's 120 characters. Oh no. Uh okay.

To be fair, the dynamic range doesn't need to say that because that is like as quiet as you can go to as loud as you can go pretty much. So just full dynamic range. So what if we do word painting? So you said full just full dynamic range? It is, really, yeah.

Okay. So we're 167 characters. Let's do homophonic polyphonic. That's 151. Um standard chamber.

And actually, just probably just a chamber choir, right? Yeah. We'll see what it does if it needs to be standard. Anything else that we could remove there other than that that comma. We're seven characters over still.

Uh and that is when it has the dashes, is that saying like do it this, then this, then this? Not necessarily, it's just that's just how it's spit it out. Okay, then say take out and dante. Take out and dante. Okay.

So now we're at 119, which is perfect. Okay. So let's go ahead and take our uh that's gonna be our song title, and we're gonna put that down here as candidate one. Our lyrics, we're gonna get rid of the bolding. We don't need it.

And cut that into here. I can tell you've done this a lot of times before. Yes. And then we're gonna take our instructions for style of music 119. Uh, you know what?

Just to be on the safe side, I'm gonna move chamber choir to the top of the instructions. Okay. Yeah, I don't want to rock a piece that is like Miss Caddy's hungry. And we're gonna create variant one. And is there anything that you would change?

What at this stage? At this stage in terms of the musical instructions. Um it depends when it uses it. I think that's a difficult one. Because dissonant harmonies do work as long as they're resolved.

If they're not resolved, if it's just close dissonant harmonies, then you don't want to listen to it. I mean, dissonance is an interesting thing to use, but it's only satisfying if it resolves, really. Okay. I I pulled out homophonics just went with polyphonic, so I did a second variation with that. Let's see what it's come up with for our variations.

Are you ready? Yeah, alright. Alrighty. That concludes part four. Thanks for tuning in.

Stay tuned for part five. See you on the next one. If you enjoyed this video, please hit the like button. Subscribe to my channel if you haven't already. And if you want to know when new videos are available, hit the bell button to be notified as soon as new content is live.