The last several years have seen AI improve by leaps and bounds. The most advanced machine learning algorithms are no longer limited to interpreting the world — they can create new content, including art so refined it can win contests and essays that can pass an MBA exam. Google is turning its attention to the musical realm. The company’s new MusicLM app can generate original music from a text prompt or a few seconds of humming.
The new AI was revealed in a paper published on the arXiv pre-print server last week. The paper explains that other teams have been able to use AI to create limited “acoustic events” over several seconds, but AudioLM takes it to the next level by generating high-fidelity audio that remains coherent over several minutes. That is, the AI output sounds like a professionally composed song rather than a collection of random snippets.
MusicLM is still just an internal research project, so you can’t play around with it yourself. However, Google has posted many examples of the AI’s compositions on GitHub. The bot can take audio input like humming a few bars of a tune, but more impressively, it can decipher a rich text prompt like “The main soundtrack of an arcade game. It is fast-paced and upbeat, with a catchy electric guitar riff. The music is repetitive and easy to remember, but with unexpected sounds, like cymbal crashes or drum rolls.”
Some of the AI-generated tracks are surprisingly listenable, but they do have a distinctively electronic vibe even when imitating other styles. Some of the tracks have vocals, consisting of mostly haunting nonsense words. It’s like a song sung in a language you nor anyone else has ever heard before.
It will take time before this tool and others like it are able to match the skill of human musicians, but that day may be coming. In order to support future research, Google has released a dataset of 5,500 music-text pairs, the same data used to train MusicLM. This will no doubt raise new concerns about the role of AI in the creation of art. Many communities and artwork repositories have had to grapple with AI-generated images, with some choosing to ban the content. Others, like Adobe, have embraced AI art with a few restrictions.
There’s also increasing concern over the copyright implications of AI. After digesting information, chatbots may regurgitate phrases that could be considered plagiarism. Music is already a fiercely litigious industry, and the use of AI-generated music could lead to lawsuits.
Now read:
- Microsoft in Talks to Invest $10 Billion in ChatGPT
- AI Art Generator DALL-E Now Available to Everyone
- Federal Circuit Court Rules AI Can’t Hold Patents
from ExtremeTechExtremeTech https://ift.tt/Lu7r45n
ليست هناك تعليقات:
إرسال تعليق