When composing a song, I find it sometimes difficult or impossible to find a style that I really want. At times, this is a benefit because the composed song moves in an unplanned, surprising and good direction. Other times, I give up and claim defeat.

I think BiaB could benefit from AI by building custom styles in real-time that meet the user's needs. One of two types of user inputs could be used to accomplish this.

1. An audio file of a song whose style you want to replicate could be imported in MP3 or WAV format.

2. A pull-down menu containing hundreds of song titles would allow you to select a specific song.

[Note that option 1 would be a much more elegant and complete solution and wouldn't require humans to continually update a song title list.]

With AI assistance, once the desired audio file is imported or song title selected, the algorthm would analyze it and return all of the instruments and other musical qualities that define that song in the form of a style. The user could then use that custom-built style just like you use any of the currently available styles; add a chord progression to it, swap out instruments, etc.

In other words, rather than scrolling thru dozens and dozens of pre-built styles and forcing one of them into the song you are creating, it would build, from scratch and return the best style that matches the song you are interested in. Icing on the cake would be for the 5 or 10 best fitting styles to be returned for you to audition.

Taken to a logical conclusion, this could eliminate pre-built styles altogether. Everytime you want to create a song, the style would be built for you in real-time based on user inputs. Another benefit might be a reduction of the footprint size of the program since the rules to create a style would replace a database of styles.

The goal here is for no one ever to "claim defeat" because they couldn't find a style that matches what they hear in their head.

No doubt, this would be a large software effort but smaller than what Microsoft has invested to create Bing. And the skeleton of an AI module to accomplish this may very well be commercially available; if not I'm sure 20-something grad students would love to work on something like this, especially if their names are associated with the end product.

On a technical note, AI, and more specifically, artificial neural networks are very good at classification. I'm thinking that this is a classification problem. So the neural net would have to be trained on several songs in every genre and sub-genre for which coverage is desired. Blues, country, hard rock, light rock, reggae, the various flavors of jazz, etc. would need to be tackled one by one.

EDIT:
Here is a list of some commercial products. I didn't realize that Microsoft offers a couple of commercial solutions.
https://www.g2.com/categories/artificial-neural-network


Last edited by Bass Thumper; 06/30/23 10:55 AM.
https://soundcloud.com/user-646279677
BiaB 2025 Windows
For me there’s no better place in the band than to have one leg in the harmony world and the other in the percussive. Thank you Paul Tutmarc and Leo Fender.