Goalthe company behind Facebook, WhatsApp, Instagram and the Metaverse Mark Zuckerberg’s are not far behind in the race and furor for AI-powered assistants and have launched their new MusicGen platform which, as the name suggests, is capable of creating music. It’s kind of like ChatGPT, but for those looking to get songs or audio samples.
The platforms of Artificial intelligence They are going through a crucial moment where they are growing by enormous steps. Helped even by humans who are about to be relieved of their jobs by these platforms.
It is in the midst of this environment that we have seen how sites of the caliber of ChatGPT itself, Midjourney and the like have been a point of controversy, both due to the lack of transparency in their regulation and the unreal efficiency with which they work, generating pieces in seconds that could take much longer for a mere human.
However, in the midst of all this panorama, there was something that remained a “safe” ground for those who felt threatened by the intrusion of AI into creative fields: music.
But the truth is that the last few months we had begun to see how these platforms also turned out to be capable of creating music and imitating the styles of artists living and dead.
Now it is taken to a new point with the creation of an AI that can create melodies from scratch, from simple text.
This is MusicGen: Meta’s Artificial Intelligence to create music that is also Open Source
He Official site where Meta has released MucisGen shows us in detail some of the products, sounds and melodies that this platform can achieve.
Along with colleagues from The Decoder They have published an interesting article where they detail in greater depth the operation of this open source project that could well have a high impact.
Thanks to AI systems such as ChatGPT, we are used to carrying out an instruction in the form of a text (or Propmt) to obtain a response or product in return.
In the case of MusicGen everything works more or less similar. Since it is possible to make prompts through text even audio files. That’s when the system basically breaks down the audio data into smaller components using Meta’s Encodec technology.
The MusicGen model was trained on 20,000 hours of music, taking as its backbone more than 10,000 high-quality audio tracks, along with music data from Shutterstock and Pond5.
YouTube begins to fill up with videos where the first enthusiasts are already showing the results of using AI. The music it generates isn’t spectacular at this early stage but it’s perfect for remixing samples and it’s a fact that it can grow a lot more.