[ad_1]
Meta CEO Mark Zuckerberg has recently announced generative AI is coming to every single one of the company’s products. Following the announcement, Meta launched MusicGen for music lovers and those interested in composing music.
MusicGen is an open-source deep-learning language model from Meta’s Audiocraft research team that can create songs using texts. The tool lets users create a song by describing the style and tone.
MusicGen is now live on Facebook’s Hugging Face AI website. You first need to describe the type of music you want, like “A cheerful country song with acoustic guitars.” Then, you can condition the music on a given melody. Finlay, click the Generate button and wait a few minutes for the outcome.
Meta introduces MusicGen, which can turn texts into music
The tool is now able to craft up to 12 seconds of music based on the input you provide to it. You can also use your own GPU or a Google Colab. Thanks to its open-source nature, MusicGen is a new playground for developers. Additionally, it can be used for generating commercial music. By facilitating more efficient and accessible music production, this ground-breaking technology is poised to transform the music business.
Meta’s Audiocraft research team uses 20,000 hours of licensed music, 10,000 high-quality music tracks from an internal dataset, and Shutterstock and Pond5 tracks to train the MusicGen. The tool employs Meta’s 32Khz EnCodec audio tokenizer for generating faster results.
MusicGen is the second AI music composer after Google’s MusicLM, which was launched last month. Hugging Face ML Engineer Ahsen Khaliq said in a tweet, “Unlike existing methods like MusicLM, MusicGen doesn’t require a self-supervised semantic representation [and has] only 50 auto-regressive steps per second of audio.”
With MusicGen, the Meta AI team hopes to increase everyone’s accessibility to music production. The art of music composition and instrument playing traditionally needs a high level of talent and knowledge. With MusicGen, people can create music using only simple text inputs.
[ad_2]
Source link