Photo : Possessed Photography - Unsplash
Artificial Intelligence

Meta has a music-generating AI: MusicGen 🤖💽

Cliquez ici pour lire en français

After text generation and image generation, now it’s time for music generation. Meta has just launched an AI capable of generating music from a short description.

From text to music 🎼

Facebook’s new AI works in the same way as other generative AIs such as ChatGPT or MidJourney. Simply describe the type of music you want in a brief sentence, and the AI will create a complete music track. As with other generative AIs, the result can be refined by specifying the musical genre sought, the instruments to be included or the emotions the track should evoke. The user can also provide a well-known melody for the AI to draw inspiration from.

Over 20,000 hours of music to train AI 🤖

In the expression « Artificial Intelligence », there’s the word « Intelligence ». However, all popular generative AIs are not really « intelligent ». They don’t create from nothing. These generative tools are designed using a huge amount of data, and we’re talking about learning. To train their AI, the Meta researchers used a Transformer-type deep learning model, designed to process large quantities of information. This model was trained with a database comprising 20,000 hours of licensed music from an in-house corpus, Shutterstock and a site like Pond5. The entire MusicGen code is freely available on the GitHub platform, allowing anyone to download and use it on their personal computer. There are four different versions of the template, suitable for different types of computer:

  • A compact version containing 300 million parameters, favoring speed and portability over precision.
  • An intermediate version with 1.5 billion parameters.
  • A large version with 3.3 billion parameters.
  • A version called Melody, with 1.5 billion parameters, which also supports audio files.

To give you an idea, the GPT-3 model used by ChatGPT has 175 billion parameters.

How can you test MusicGen? 💽

If you’d like to try out MusicGen without having to download the source code, it couldn’t be easier. Meta has posted a free demo on the Hugging Faces platform. Once on the platform, all you have to do is describe the music you want to produce. The AI understands English and French, and will generate a 12-second track. The clearer and more precise your request, the better the quality of the result.

What do you think of generative AI and its impact on our daily lives? Tell us in the comments section.


Source : 01Net

Qu'en avez-vous pensé?

Je suis fan
Je me questionne

Vous pourriez aussi aimer

Laisser une réponse

Votre adresse e-mail ne sera pas publiée. Les champs obligatoires sont indiqués avec *