Is Google’s music AI a (new) threat to artists?

Zilber

New member
After the successes in image (Dall-E, Stable Diffusion) or text (Chat GPT), AI is demonstrating its creative power in music through Google’s MusicLM. But Google remains cautious and does not give access to its tool to the general public. For fear of opening Pandora’s box?



Images, then text and conversation, and now music! The demonstrations of AI models follow each other and become more complex and today Google’s MusicLM model is in the spotlight. An AI that, like its sisters Dall-E and Stable Diffusion for the image, is based on a request in text form to compose pieces of music for you. If such models have been designed in the past, here the “high-fidelity” quality takes a leap forward in terms of display. And also credibility and diversity. On this page dedicated to MusicLM, Google researchers share many musical fragments with corresponding searches. In addition to the variety of styles, it is necessary to underline the nature of the requests: short pieces, long pieces, segmented by genres (Story Mode) or even variations on well-known melodies or compositions inspired by descriptions of famous paintings.

Taking a moment to listen to several of these “creations” makes a few observations seem obvious. On the one hand, certain pieces, especially electronic ones, are clearly at the level of human productions. If the AI is then still able to make “loaves” on the piano (see ” Text and melody conditioning », column « bella ciao – to hum “, line “piano solo”), great progress has been made in the construction and in the generation of the voice (particularly in terms of timbre). If the model makes a big deal out of not sticking to one language, then the example is ” Slow reggae song led by bass and drums. Sustained electric guitar. High bongos with ringtones. The vocals are relaxed with a relaxed feel, very expressive. “in the section” Audio generation from rich captions is quite impressive. From a long, precise description, the AI manages to make a really believable piece with human vocals in it. An example that, in our opinion, illustrates one of the reasons why, unlike the AIs already mentioned, for the time being there is no possibility to “play” with this AI.

A potential earthquake for the music market?

If not all pieces are convincing, however, we feel the power of Google's AI in certain examples of musical creations.

If not all pieces are convincing, however, we feel the power of Google’s AI in certain examples of musical creations.

A front of artists is emerging in the field of drawing, the designers of AIs such as Dall-E and Stable Diffusion have indeed trained their precious algorithms on entire corpus of creations of artists who are still alive. All this to enable a quidam – like you and me – to convey in two sentences the style of an artist who has sometimes spent his life perfecting not only his art, but also his visual identity. Between this move that could trigger a serious legal correction, and the example of Chat-GPT where examples of plagiarism, cheating and other school “DIY” explode, the arrival of a powerful audio tool to force Google to (good) cool head.

Read also: This free AI has only one goal: to hack copyrighted images (January 2023)​

Because the field of possibilities and thus the potential abuses is just as enormous in the musical field as it is in the visual field. In addition to the weakening (or even the destruction) of markets for the creation of sound identities, musical backgrounds, background music, etc., the risk of plagiarism or “strong inspiration” that could arouse the wrath of one or more well-known artists is potentially high. And if OpenAI was ready to “disrupt the market” with Dall-E and Chat-GPT to attract investors (like Microsoft), then a behemoth like Google has a lot more responsibilities and pressure on its shoulders. A responsibility that is clearly understood and explained in the research article “Music LM: generating music from text” (article in English): ” We strongly emphasize the need for additional future work to address (the) risks associated with music generation – we do not intend to release any models at this stage ». In short: Google doesn’t want to open Pandora’s box. Just a smaller player with less risk “commercial” could afford it…

Read also: “There is nothing revolutionary about it”: French AI pioneer Yann LeCun is not impressed with ChatGPT (January 2023)​

And someone will logically do it! With the mass of corporate publications such as Google, many musical models are now the subject of intensive research. If the threats are important to the musical ecosystem, the promises are just as important to the rest of the world. Be it the ability for a company to create unique, royalty-free music on hold, the quick and easy generation of sounds and music for an independent video game, etc. AI will obviously shake up the music creation market. The question is whether a legal framework can regulate its arrival or whether it will act as a new tidal wave. On a segment that has already known many.

Source :

The edge
 
Top