Meta has released an open-source version of its AI music generation model, called MusicGen, which works similarly to ChatGPT in generating text. The model uses an EnCodec audio tokenizer based on a transformer language model. Users can demo MusicGen through Hugging Face’s API, which may take some time depending on the number of users. Otherwise, users can download and run the code if they have the know-how and a supportive rig.
The system can take existing genres or songs and modify them, such as creating an 80s pop song from a classical musical refrain. Our own tests revealed a synth-heavy symphonic happy birthday rendition and a lo-fi hip-hop track containing samples of nature, including crickets. Although the AI can generate complex musical scores, it may not comprehend or replicate certain composers’ musical styles.
Meta claims that MusicGen’s training data used to develop the algorithm includes 20,000 hours of licensed music from an internal dataset comprising 10,000 music tracks. They also used approximately 390,000 instrument-only tracks featured on Shutterstock and Pond5. However, artists have criticized the use of their work to train AI models without their permission. Some artists have sued the biggest AI art companies like Stability AI and Midjourney.
Meta has joined in on AI developments, stating that it wants to release more open-source models into the ether that anyone can use. However, creatives remain concerned that companies will use AI for artistic tasks instead of real flesh and blood creatives. The AI can represent an unfair competition for artists, according to a Meta researcher. Nonetheless, the open models provide new tools for music amateurs and professionals to use.
Frequently Asked Questions (FAQs) Related to the Above News
What is MusicGen?
MusicGen is an open-source AI music generation model released by Meta that can generate complex musical scores based on existing genres or songs.
How does MusicGen work?
MusicGen works similarly to ChatGPT in generating text. It uses an EnCodec audio tokenizer based on a transformer language model.
Can users demo MusicGen?
Yes, users can demo MusicGen through Hugging Face's API, but it may take some time depending on the number of users.
Can users download and run MusicGen?
Yes, users can download and run MusicGen if they have the know-how and a supportive rig.
What kind of music can MusicGen generate?
MusicGen can modify existing genres or songs, such as creating an 80s pop song from a classical musical refrain. It can also generate complex musical scores, but it may not comprehend or replicate certain composers' musical styles.
What kind of training data did Meta use to develop MusicGen?
Meta claims that MusicGen's training data includes 20,000 hours of licensed music from an internal dataset comprising 10,000 music tracks and approximately 390,000 instrument-only tracks featured on Shutterstock and Pond5.
Have artists criticized the use of their work to train AI models without permission?
Yes, some artists have criticized the use of their work to train AI models without their permission and have sued AI art companies like Stability AI and Midjourney.
How does Meta plan to use AI developments in the future?
Meta wants to release more open-source models into the ether that anyone can use.
Are creatives concerned that companies will use AI for artistic tasks instead of real flesh and blood creatives?
Yes, creatives remain concerned that companies will use AI for artistic tasks instead of real flesh and blood creatives. The AI can represent an unfair competition for artists, according to a Meta researcher.
Who can use MusicGen?
MusicGen provides new tools for music amateurs and professionals to use.
Please note that the FAQs provided on this page are based on the news article published. While we strive to provide accurate and up-to-date information, it is always recommended to consult relevant authorities or professionals before making any decisions or taking action based on the FAQs or the news article.
