Meta Unveils Open Source AI Music Generator

MusicGen - Meta's Open Source AI Music Generator

Meta, the social media giant, has developed a music-generating tool called MusicGen. The tool can create audio based on a text description, such as “An ’80s driving pop song with heavy drums and synth pads in the background.” A demo of MusicGen is available online and can produce approximately 12 seconds of audio. A new feature has been added to MusicGen that allows users to steer the software with reference audio. By using an existing song, MusicGen will attempt to follow both the description and melody of the reference audio. This optional feature provides users with more control over the music generation process.

According to Meta, their latest creation, MusicGen, has been trained on an extensive collection of music. The training dataset includes 20,000 hours of music, consisting of 10,000 licensed music tracks of high quality and 390,000 instrument-only tracks sourced from Shutterstock and Pond5, which are both well-known stock media libraries. In a recent development, the company has made pre-trained models available for public use. However, the code used to train these models has not been provided by the company. Interested users must have a GPU with approximately 16GB of memory to run these pre-trained models.

Source: Twitter

MusicGen’s performance was recently evaluated. In a recent statement, the speaker expressed their opinion that while technology has advanced to a point where it can create music, it is not yet at a level where it can replace human musicians. A recent review has praised the melodic quality of a certain type of music. The music in question falls under the category of “ambient chiptunes music” and has been deemed reasonably melodic by the reviewer. In fact, the reviewer goes as far as to say that the music is on par, if not slightly better, than the results from Google’s AI music generator, MusicLM. The team in question is unlikely to receive any awards.

The use of generative AI to create homemade tracks that mimic familiar sounds has been gaining popularity and going viral. These tracks are able to pass off as authentic or at least very close to the real thing. In recent news, music labels have been swift to notify streaming partners of copyright infringement concerns, and have largely emerged as the winners in these disputes. The issue of whether “deepfake” music infringes on the copyright of artists, labels, and other rights holders remains unclear.

It might not be too long before there are rules to follow. Several lawsuits that are currently going through the courts will probably have an effect on AI that makes music. One of these lawsuits is about the rights of artists whose work is used to teach AI systems without their knowledge or permission.

Meta, which is not placing any limitations on how MusicGen can be used, claims that all of the music used to train MusicGen was “covered by legal agreements with the right holders,” such as a deal with Shutterstock.

Read Also: Meta Introduces AI Sandbox For Advertisers

PTA Taxes Portal

Find PTA Taxes on All Phones on a Single Page using the PhoneWorld PTA Taxes Portal

Explore NowFollow us on Google News!

Leave a Reply

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

Back to top button
Get Alerts!

PhoneWorld Logo

Join the groups below to get the latest updates!

💼PTA Tax Updates
💬WhatsApp Channel