AI & Music – Where It Is and Where It Should Go (but probably won’t…)

As a creator and someone that works in technology (beats don’t pay all the bills fam…), I’ve been watching the mash-up of newly emerging AI technology and music closely. Most of what I’ve seen has either been a “fun parlor trick” or downright terrifying. For those following less closely, I wanted to discuss the current state of AI & Music and where I think it should go.

What’s Happening Now?

Generate Music in the Style of… The first time I heard of AI as it relates to music, was a YouTube site where the first few bars of a popular song (ex. Nirvana’s “Teen Spirit”) would be continued by AI. The results weren’t exactly listenable (gibberish lyrics, sound artifacts) but at the same time were better than anything I could have imagined. As technology progresses, I have no doubt in the next few years there will be an app that allows you to plug in any popular artist/theme and generate your own custom song.

PROMPT: Generate a song by the Beatles about the series finale of Stranger Things.

IMO, this is the most dangerous type of AI use within music – being able to replicate an entire artist’s sound, vibe, aesthetic, themes, etc. (and most likely no compensation). I also feel it will lead to a dystopian music scene where everything is regurgitated older music and actual musicians/composers will become less and less relevant.

Vocal Replacement – This is what I refer to as a “parlor trick”. You have heard of the fake Drake song generated by AI or Notorious B.I.G. rapping Raekwon’s verse on the Wu-Tang classic “C.R.E.A.M.“. From what I understand, AI gets trained on an artist’s voice, tone, cadence, inflection, etc. using their accapellas. A “source” verse is fed into AI and it replaces the voice with the trained AI version. This is the lamest use of music related AI (in this context, but more on that later…). I don’t understand the “wow factor” of essentially Biggie doing Wu-Tang karaoke. The fake song variant is even worse, as the original artist (at this time) isn’t being compensated.

Chord/Melody Creation – This is actually one current use I don’t have too much issue with. Creative block is real and getting some foundational music theory help isn’t wrong. There is no difference between watching a YouTube video about Jazz chord progressions vs. asking AI to give you some common voicings. Where I do take issue is artists being lazy and using out of the box musical ideas without adding anything to them. However, these artists will (and have always) existed with or without AI in the mix.

Where Should it Go?

Mixing – Chill out engineers, I’m not going after you. What I would like to see from AI is direction on how to get a specific sound. For example, I create my song and I want a similar studio sound to 1960’s Motown. If AI is trained properly, it could provide direction such…

Motown vocals are typically between frequencies ### and ###.  Your vocals are in the range of ### and ###.  Consider adding eq to cut frequency in this range.

Motown drums contain reverb at % amount.  Consider adding plate reverb in this  X range.

Vocal Replacement “Hold up, didn’t you just say…” context is key here! When the original artist is involved, I think this type of technology is fine. Look at artists that have had their voices permanently damaged – D.O.C., Beanie Sigel, etc. They could still record their verses with current voices and have it replaced to sound like them prior to damage.

Restoration – This is another one. Generally I’m not a fan of “Re-mastered” music, but I could see AI being used to restore old, damaged recordings and demos. For example, I imagine what the Wu-Tang demos would sound like if AI replaced the verses with clearer vocals and removed tape hiss.

Why It Probably Won’t Happen

Money – that’s it and pretty sure we all know it. I foresee a new licensing model where artists can grant permission to use their vocals etc. to generate new songs. Albums sales and streams won’t make money – licensing your voice will.

Record Labels will also create entirely new artists with auto-generated songs. They can feed AI with every hit song ever created to get a generic “hit formula”.

Eventually, new artists will only create music as a hobby and AI will have no new music to learn from. Music will become one homogeneous genre of perfectly quantized drums, proven chord progressions/melodies, and lyrics geared to appealing to the largest amount of people.

Or maybe I’m all doom and gloom and wrong :shrug: