I wonder how well it would work to use AI as a front end to Band-in-a-Box?
Band-in-a-Box is a commercial program that has been around since 1990. What it did then was let you specify a chord progression, style, tempo, and instruments and it would make a generate a MIDI track. I think it might have also been able to take a melody and come up with a chord progression for it in a style/genre of your choosing.
The target market was musicians. Instrumentalists used it generate tracks to improvise or solo with for example, and songwriters found it useful to essentially have a full band at their beck and call while composing.
Over the years they added more features, and switched to sounds from recordings of real instruments played by real musicians. They have very good stretching and pitch transposition so you can use these at a range of tempos and keys and they still sound good.
It is still aimed at musicians, and can be overwhelming to others. This I've read is made worse because as it has grown in features and capabilities in the 25+ years it has been available the interface has become kind of disjoint.
It is not something the kind of person who just wants to describe what they want to hear and have a song produced would enjoy. But if an AI could operate it for them, maybe that would work and the result would be something with much better sounding instruments than the AI song makers (and without the risk of including unlicensed copyrighted material).
BIAB is still best in class (even if the UI is practically Soviet era) simply because of the sheer number of RealTracks, which are actual performances by musicians that dynamically adapt to your chord progression.
I’ve actually taken some of my own compositions and run them through Suno using the “Cover” option, and it’s pretty nuts what it can do.
What would be really cool is the concept of combining a physical arranger keyboard (like a Yamaha PSR-SX) with real-time orchestration produced by a backing generative model.
https://mordenstar.com/blog/dutyfree-shop
> This I've read is made worse because as it has grown in features and capabilities in the 25+ years it has been available the interface has become kind of disjoint.
It's impossible to exaggerate how true this is. I often say "BiaB is the best worst software - or should that be 'worst best software'? - I've ever used." A toolbar that crams dozens of tiny icons, almost no visual hierarchy, dated visual style, waaaay too many dialogs (dialogs within dialogs!), zero discoverability, inconsistent labeling, basic features missing...I could go on. To add insult to injury, I'm using the Mac version and it looks/feels like a port, not a native app.
I like the direction Apple is taking with their digital audio workstation, Logic Pro X. While not overtly AI, they've been introducing intelligent musical features starting with their Drummer feature several years before AI became commonplace.
These days I'm programming drums on dedicated hardware but Logic Pro's Drummer feature had been immensely helpful for me as a guitarist who hadn't done much drum programming but wanted to play along with interesting drum beats while arranging a song. Just a few options but that's what makes it so approachable. It's helped me keep the song "mine" without the hassle of sourcing loops/samples manually, even if only temporarily.
I remember watching a youtube video that was kind of a Star Wars fan fic. It had a great soundtrack, that was a cross between John Williams and Michael Giacchino. The YouTuber was using some commercial program that included samples of all the orchestral instruments and you could use it to compose lush scores. I never used it since it was expensive, but I always dreamed of tools like that, like GarageBand on steroids for orchestras. Now I wonder how quickly I could vibe code something like that...
The code is only a (very important) part of this type of program. The samples are critical and (for the time being anyway) can't be generated by AI.
Especially important if you want orchestral instruments that sound realistic. Just think of the many ways that a single note can be played by a professional player and multiply that by the range of the instrument.
Edited to add: not orchestral instruments, and also not samples, but this gives an idea of the complexities of capturing the characteristics of an amplifier so that it can be modeled faithfully: https://neuraldsp.com/quad-cortex-updates/introducing-tina (I'm not related and I'm actually a Line6 customer, but I saw this at work in an interview by Rick Beato and though it was super interesting)
Is this the vid? https://www.youtube.com/watch?v=9YL8pwF7Mnc
Rick Beato travels to NeuralDSP in Finland.
Agree 100%. The multivariate ways a note can be expressed is almost unlimited. For example, I first heard Bach's Cello Suite #1 played by some random cellist. Fell in love with it and listened to it endlessly. Then I heard Yo-Yo Ma play it and it was a completely different piece.
IIRC the samples in this program were actual performances, so I'm curious how they captured all the variations...
There is a whole world of expensive virtual samples instruments that can very convincingly replicate an orchestral performance in a DAW. See Spitfire Audio, EastWest, Cinesamples, etc.
> I wonder how well it would work to use AI as a front end to Band-in-a-Box?
Wow, I haven't heard that name since... well, since the software was relatively new.
I do like the idea of an AI music tool that lets you have that kind of workflow, choosing a level of granularity (and, presumably, being able to edit the intermediate results etc.).
Personally I would feel ripped off if I'd bought music that would be generated by AI but there was no notification that this was done. Any group that would do so I would permanently disregard for any further consumption or attention from me.
this is an overcorrection not understanding the extent to which music is the product of shared rules that are more an act of mechanical execution than creativity. the creativity is there it's just much smaller than many realize. machine generated music has existed for a long time