Cover Songs & AI
I’m “dropping” three new cover-song tracks, with a short and snarky chat about my use of Artificial Intelligence tools in the creative process. Music first!!
GHOST RIDERS IN THE SKY
Original version
My cover sketch
?
Many thanks to my brother-in-law Mike McColl on lead vocals, rocking the Johnny Cash vibe. I tried to imitate the Cash version closely, with a few add-ons for ghostly aura. The female backup vocals were AI-generated, and I used an AI “assistant” to mimic the drum part of the original.
?
I WANT CANDY
Original version
?
My cover sketch
I attempted the “swingin’ 60’s” rhythms of the original, but I couldn’t quite get that groovy, so I took a cue from the 1980’s “Bow Wow Wow” cover, pushing the drums to a more electronic on-the-beat feel and imitating the signature guitar riff of that version. That carried me to the short EDM-like breakdown/buildup section — which I really like. AI-tools were my able-assistant in many signal processing tasks, but provided nothing “generative” for this track.
CRIMSON AND CLOVER
Original version
My cover sketch
?
I wanted to do this song because, to put it in business terms … it is really sexy (that’s OK to say LinkedIn, right?). Pulsing tremolo, suggestive tempo increase/decrease, seductive lyrics. This crazy simple song still holds up after oh these many decades. Like “Ghost Riders”, I aimed for a pretty straightforward imitation of the original. Where did AI come into the picture?? Mainly the female backup vocals (plus the usual signal processing). The guitars on this were an interesting challenge and forced me to some creative solutions with the sample-based instrument I used (see below).
?
AI Tools Used in Composition
No AI's were harmed in the production of this music.
What did they do?
Bing Copilot . ?Yep – My “Everyday AI companion” helped me find lyrics quickly and answered a few rando question about drum beats and tremolo.? Per the instructions (“Copilot uses AI. Check for mistakes.”), I checked for mistakes. Sidenote: that little disclaimer got me thinking:? maybe we should label everything made by humans, “Made by humans. Check for mistakes.”? I mean … it’s valid, right? One day, the AI Benign Benevolent Authority (ABBA) will require this of us.
Sidenote #2: Copilot keeps getting amazingly better. I mean … really, really better. I’ve asked wonky questions and it understood what I was asking, produced meaningful answers with solid reference links, and been able to drill down into follow-up questions keeping context of the discussion.? I continue to be blown away.? As for writing lyrics – yeah, it’s pretty good.? A little annoyingly good, actually.? But … I like writing my own lyrics and I’m not songwriting as a profession, so I don’t care.
Emvoice .? I enter MIDI notes and lyrics. Emvoice produces a sort-of-likeness of a human being singing those lyrics and notes. Pretty cool.? I mainly used this for background vocals. On a lead vocal its … how shall we say? Taylor Swift need not worry.?The Beastie Boys need not worry. See my mildly snarky review at bottom of this post. It was a useful tool for me to create backing vocal harmonies.? I wouldn’t use it for a lead vocal on the type of music I did for these covers – they need the organic sound of a real human voice.? But on an EDM track?? Yeah, this might work as lead vocal on a heavy-synth dance track, or atmospheric electronic piece.
KONTAKT . This is a flat-out amazing piece of software. But best I can tell, it doesn’t use AI techniques … yet. I was mainly using the “Session Guitarist” libraries. These are collections of human-played guitar parts – strums, melodies, scratches, slides, bends, slaps, endings – you name it. The software allows you almost limitless ability to combine those and pitch-shift them to produce really convincingly real guitar performances. Under the hood, it is real guitar performances.? But the ability to smoothly move between all the variations and make it sound like someone played the part you (as the composer) wanted … totally incredible.? This was my big learning curve for this project. Totally fun.
What I want to see next in KONTAKT is the ability to describe what I want in natural language and have the tool attempt to render a performance – just the way an actual session player would do.? As it stands today, I still need to poke the keyboard, turn knobs and levers, tweak parameters in the UI.? All great fun, of course, but it would be great to just give it verbal instructions and have it play the guitar.
Izotope RX10 + LogicProX Drum Doubler . RX10 is audio voodoo. I don't know if it uses AI or what, but it can extract the entire drum part (kit, percussion, etc.) from a fully mixed track. That's crazy-town signal processing. Lots of weird Fast Fourier Transform stuff going on, I assume. Logic Drum Doubler can then listen to that extracted part and generate MIDI from it. Again ... signal processing magic, not necessarily "AI" methods, per se. But at the end of it, you have a MIDI-encoded version of the original drum track (with some errors, of course) as a great starting point for customizing. It's so fun.
From there, it's a bit manual to assign different drum set sounds to the MIDI. What would be a really cool AI addition? I would love it if there was an integrated AI tool in Logic that could listen to the extracted track and give me the top three drum kits that would best imitate the original sound. How hard could that be?
I heard one of the Pixar writer/directors say that telling engineers a thing is impossible is the best way to get it done. So, I say to you engineers at Apple who build Logic Pro: it can't be done! Ha! Just try! You can't possibly make a tool that could listen to a mixed track and tell me the best drum kit to imitate the original sound and assign the MIDI notes properly to the parts! I dare you! I double-dog dare you!
Some Thoughts on AI and the Joy of Creating
Big Up-Front Caveat:
My comments are made in the context of a music hobbyist.? I suspect that someone who makes their living in music production in any capacity would have a different view.? If they are a big farm, I am just a guy with a small garden in his backyard.? Different motives, means and opportunities. / Caveat finished /
I’ve commented in prior articles on AI and music ( Part 1 ? /? Part 2 ? /? Part 3 / Part 4 ) that when it comes to how AI interacts with humans, I like Microsoft’s vision of AI as a complementary assistant – a helper of human intelligence rather than a replacement of it. That’s why Microsoft calls it “Copilot” rather than “Autopilot”.
Will AI replace some tasks we might call “intellectual drudgery”, similar to automated machines replacing repetitive human labor (e.g. – putting bolts on wheels in a car-manufacturing facility)? For sure.? It’s already been happening for decades. Will the new wave of “generative AI”, with its ability to read and summarize documents, create images from text descriptions, see pictures and listen to voices -- will it take this to new levels? Definitely.
But does that necessarily mean it will replace human creation? Is there a day when there are no more musicians, illustrators, or writers?? Is the sky falling? I kind of doubt it.? Why?? Two reasons.? First, because of Joy.? The AI tools I use in music creation are incredible. They save me time, they provoke new ideas, they help me do tasks I don’t really have the expertise to do well. But one thing they don’t do is ENJOY the act of creating. Only I do that … because I am human. As long as humans really enjoy doing something, they are going to do it, and they will have fun doing it. One point of reference: Society for Creative Anachronism ( sca.org )
Secondly, humans are absurdly competitive (not a quality I always love), which often inspires amazing creativity when people aren’t just being stupid. So, the AI ups the game on interesting music, writing, images, architecture?? Bring it on!? Humans will make more interesting music, more clever writing, more emotional images and even better architecture! Or, we will take a total 90-degree turn where the AI wouldn’t even think to do so. Blah to you, AI!? You suck, we’re awesome! Now give me a summary of the emails I missed while writing this piece. And order me a pizza. You know the kind I like.
Look, let’s be honest – No one really knows what’s about to happen. We’re opening Pandora’s Boxes left and right. It’s gonna' get weird. Maybe we’ll survive. What do I know? I’m just a guy poking around in his music garden.
Geoff’s mildly snarky review of Emvoice
note - this is based on the verbatim text on the Emvoice website
Claim: “In simple terms, it’s a state-of-the-art AI-powered tool that allows you to generate realistic and high-quality vocal performances.”
Geoff’s Response: Does it sound like Adele? No.? Does it sound as real as me singing it? No.? Is it more in tune that I am? Definitely. Can I sing male and female parts that are well outside my range? Bingo. That’s why I use it. But "realistic"? That's a stretch.
Claim: Natural-sounding Vocal Synthesis. Emvoice uses advanced AI algorithms to produce natural-sounding vocal performance. Say goodbye to the days of robotic and artificial voices. You can create vocals that are virtually indistinguishable from human recordings.
Geoff’s Response: Not so fast there, Skippy! My experience was a bit of a slog through the trough of disillusionment . For some ideal combinations of melodic phrases and lyrics it’s passable.? Wander much outside of those and you know it’s a synth. For buried background vocals, you can get away with it. Want replicate the singers on Steely Dan album? Hire real singers – good ones.
Claim: Real-time performance control: One of the standout features of Emvoice is its real-time performance control. With this tool, you have the ability to manipulate various aspects of the vocal performance, such as dynamics, pitch, and vibrato.
Geoff’s Response: Totally NOT real-time.? Pre-composed, yes. Decent quality?? Eh … See above.
Claim: Integration with popular DAWs: Emvoice seamlessly integrates with popular Digital Audio Workstations (DAWs), making it a breeze to incorporate the tool into your existing workflow. The smooth integration ensures that you can focus on your creativity rather than technical hurdles.
Geoff’s Response: Let us consider the terms “seamlessly” and “smooth” – this should mean, literally: I cannot see the seam. It’s smooth, like one piece of cloth.? Having worked in software sales for many years, I myself have been guilty of uttering these words (though I try to qualify them!). ?Seamless … is sometimes true.? More often, it is a nice, clean, tight seam … but seam it remains.? As for Emvoice and my DAW (Logic Pro), it is a seam you can put your foot through. I won’t trouble you, dear reader, with the frustrating details. Let us just say, “usable it is; seamless and smooth it is not”. I look forward to improvements in this part of the product.
Now that I’m done, if you want a real review, see here: Review: Emvoice One Vocal Synthesis ( mixdownmag.com.au )