Meta has launched an AI model that specializes in music generation

Another solution developed and announced by Meta, involves artificial intelligence (AI). The novelty of that time would allow users to use MusicGen, a template that makes it possible to create songs from text commands. The feature can also be used on songs that have already been created.

Read more: Bard vs ChatGPT: Google’s AI wins the fight with 4 free jobs

According to the Olhar Digital portal, the model has received trainings containing 20,000 hours of music, as well as 10,000 audio clips. Meta states that the program It offers high quality samples, following a consistent structure, based on a description given by users. Check out more details throughout the text.

The training also included 390,000 musical instruments from Shutterstock and Pond5. This model is currently open sourced github. With the novelty, the user can abuse their imagination and creativity to generate songs, bringing out all the elements, such as style, instrumentation, and melody.

How does Meta MusicGen work?

Meta indicates that the creation received more satisfactory results than Google’s MusicLM. On Twitter, Meta AI research engineer Felix Kroc revealed details of how the system works. He shows that the source is simple and can provide answers based on the texts and melody. “We have released (MIT) code and models (CC-BY NC) for open research, reproduction and for the music community,” noted the expert.

The engineer also explains that the EnCodec audio token, as well as the Hugging Face API, were used in the development. The inclusion of this API is what allows users to test freshness online. In addition, the company says that three versions of the resource have been tested.

With this, it was discovered that the intermediate form was the most acceptable.

The latest big tech feature is a strong competitor to Google, which launched MusicLM. The market is so competitive, Meta even compared the model to Riffusion and Mousai, with similar performance.