Connect with us


OpenAI’s Jukebox model can write songs in the style of your favorite artists

Anyone for a track that sounds like Creed?

Ok, so there isn't actually a robot playing the instruments. (Image: Unsplash)

Today marks the launch of the incredibly impressive Jukebox generative model from the San Francisco-based research lab, OpenAI. It’s incredibly impressive as the team has managed to develop a system of AI that can generate music AND lyrics.

Not only that, but it can “sing” its ditties to you as well. If you have got anything to do today, I suggest you cancel it, because I’m about to expose you to a time sponge of epic proportions. We’ve seen AI generating endless melodies, before, so let’s take a look at what OpenAI’s Jukebox model can do, then.

What is OpenAI?

As mentioned, OpenAI is a research laboratory based out in San Francisco. They’ve hit the news plenty of times, be that for their DOTA-pro smashing 1v1 AI, or their hyper-dextrous robot hand. The 100-strong team continues to push the boundaries of what their AI is capable of. If it is capable of learning how to play a game, from scratch, simply by playing the game itself, then that gets filed under “suitably impressed”. That’s right, it learns in just the same way as you or me. The only difference being (I’d suggest) is that it can’t forget. Nor will it panic in crucial situations.

These are just a couple of examples of the projects OpenAI is developing. The laboratory, which has counted Elon Musk among its previous board members, has a full list on their website of the projects it is currently undertaking. You can browse those, here. Don’t say I didn’t warn you…

Put another dime in the Jukebox, baby

openai jukebox

Screenshot of OpenAI’s jukebox

I have to say, I wasn’t entirely convinced by Jukebox before I actually checked it out, but it really is an impressive model indeed. Essentially, feeding the AI with some music from a particular artist (under various conditions) allows the Jukebox AI to generate a unique track using sampling. This means the music not only sounds like the artist in question but so does the vocal.

I’ve spent some time going through some of the tracks available and, while they might not be perfect, watching them with the lyrics is awesome. The Jukebox AI returns some astonishingly good tracks (I have to admit to pissing my pants over the Creed-style track the AI has produced), with the lyrics sounding really clear and making sense.

I also noticed how someone at OpenAI has an excellent taste in underground hip-hop, with 8Ball & MJG, and Aesop Rock all appearing in the sample songs available to listen to. There is a range of styles to listen to, as well, with progressive death metal, southern hip hop, and country gospel all getting a look in. Honestly, this is well worthing losing yourself in. I particularly enjoyed the Ella Fitzgerald re-renditions.

The future of songwriting?

OpenAI has broken the mold with their Jukebox generative model. I, for one, am excited to see how this develops as Jukebox learns and adapts to the process. This could mean we’ll be hearing brand new songs from artists long since passed.

But is this the future of songwriting? I, for one, don’t believe that a machine can ever replicate the emotion a song contains. Not perfectly, anyway. If the machines start to feel, that’s probably when we need to begin worrying or maybe just living out our 14-year-old-self’s Outer Limits fantasies. Aaaaanyway…Think this is cool? Check out this scientist, who programmed his robotic arm to manipulate his synthesizers.

What do you think? Which bands or artists would you like to see Jukebox cover? Let us know down below in the comments or carry the discussion over to our Twitter or Facebook.

Editors’ Recommendations:

Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

More in News