-
-
01-09-2024 06:40 AM
-
Have you tried it? And, no good at all?
-
At last! No more stress. I can just click a button, and there it is! A REAL jazz solo. We’ve all been suckers all these years.
Joking aside, I wonder whose licks and solos were input as role models for the Ai to learn, and should those guys sue?
-
Originally Posted by hotpepper01
-
Don't know what to say after hearing that demo. Couldn't they have used a "modeled" Tele? Would have sounded so much better.
-
I need a dislike button just for this.
-
"No computer stands in my way..."
-
How do you know we've reached "the singularity"? When you tell an AI "don't quit your day job."
-
A professor from UCSC has been working on getting computers to generate music in the styles of Bach, and Mozart, for over a decade.
I heard an interview on NPR with him… maybe 8 years ago. They played some of the computer generated stuff, and it was tricky for me not to think it was the fore mentioned composers.
IDK about you, but I am not worried. We still use paper for books. We still listen to radio. Things changes, but old technology sticks around. I guess the change would be seeing ourselves as “old technical”. That kinda happens as you age anyway.
-
Ffs
-
Sounds like a "jazzy" demo song on a mini keyboard circa 1985.
-
This is not so new. Bob Keller, the creator of Impro-Visor (a kind of open-source BIAB for the poor), who died four years ago, was working on such stuff already a few years ago.
(BTW Impro-Visor can create "LEGO brick" road maps à la Cork/Elliot)
And I am still waiting for the Barry Harris sixth diminished accompaniment algorithm with an adjustable degree of borrowed notes -- can't be so difficult to program ...
-
-
I bet if they took the time to use higher quality synth or have someone perform these solos they would sound better.
What's interesting is that lately people are trying to apply newer ML algorithms to music. IMO the quality of the output has taken a step back from prior generations of generative music. Unfortunately the researchers working on this have no (or at least poor and unrefined) musical training and therefore they are unqualified to evaluate their own results.
Case in point, someone was sufficiently impressed with themselves to post this trash:
https://youtube.com/watch?v=c6DK8IKS9z4
Why do I say this is a step back? Newer ML algorithms are attractive because they require much less expert intervention/manual programming. This cost efficiency is why they are everywhere and so attractive for business applications (you can hire 3 engineers to automate what previously took a 50 person multidisiplinary organization to run). However, the tedious manual work of earlier generations of computational music really paid off. Take a look at the work of David Cope. No it's not something I'd want to listen to start to finish, but it's much better than the garbage being put out by tech giants today:
The melodies aren't great but notice the use of repetition, a definite musical form with motifs developing over time, symmetry of phrases.
Another one:
Epiphone Zephyr Regent Reissue, 2004 MIK Sunburst
Today, 08:03 AM in For Sale