This short multi-speaker transcript (from a LARRUZ YouTube video about OpenAI's MuseNet) focuses on using continuation-style neural models—like GPT-2 applied to music—to continue short musical seeds and blend genres. The speakers introduce the idea, point to audio demos (including a two-hour concert and an online demo link), highlight strengths (genre transfer, interesting continuations) and weaknesses (instrument pairing issues, limited training data), and raise copyright/ownership questions; source code availability is noted as current or imminent.
Chapters
Dear fellow scholars, this is two minute.
Papers with not so long ago, OpenAI.
Has released GPT2, an AI that was.
Trained to look at a piece of.
Text and perform common natural language processing operations on it. For instance answering questions, summarization and more.
But today we are going to be laser focused on only one of those.
Tasks and that task is continuation where.
We give an AI a bunch of.
Text and we ask it to continue it. However, as these learning algorithms are quite general by design, here comes the twist.
Who said that this can only work for text? Why not try it on composing music?
So let's have a look at some.
Results where only the first six notes.
Were given from a song and the AI was asked to continue it.
Sam.
Love it.
This is a great testament to the.
Power of general learning algorithms.
As you've heard, this works great for.
A variety of different genres as well.
And not only that, but it can.
Also create really cool blends between genres. Listen as the AI starts out from the first six notes of a Chopin.
Piece and transitions into a pop style with a bunch of different instruments entering.
A few seconds in.
And great news because if you look here, we can try our own combinations through an online demo as well. On the left side we can specify and here the short input sample and ask for a variety of different styles for the continuation.
It is amazing fun.
Try it. I've put a link in the video description. I was particularly impressed with this combination.
Really cool.
Now this algorithm is also not without.
Limitations as it has difficulties pairing instruments.
That either don't go too well together.
Or there is lacking training data on.
How they should sound together. The source code is also either already.
Available as of the publishing of this.
Video or will be available soon.
If so, I will come back and.
Update the video description with the link.
OpenAI has also published an almost two.
Hour concert with tons of different genres.
So make sure to head to the.
Video description and check it out yourself. I think these techniques are either already.
So powerful or will soon be powerful.
Enough to raise important copyright questions and will need plenty of discussions on who.
