Google Just Developed An AI Model That Creates Music From Text Prompts — But The Company Won't Be Releasing It Anytime Soon

• Google announced the creation of an artificial intelligence bot that creates music based on text descriptions.
• Due to technical issues and risks, the technology will not be published.

At the AI ​​Race, Google announced the development of a bot that creates music based on text prompts, but don't expect to be able to use it anytime soon.

In a paper published Thursday, Google researchers describe MusicLM as "a model that creates high-fidelity music from textual representations, such as a relaxing violin melody with distorted guitar."

"We show that MusicLM can be adapted to text and melody as it can transform whistle and hum melodies according to the method described in the text heading."

According to the study, users can enter descriptions such as "memorable jazz song with an unforgettable saxophone and lead singer" or "90s Berlin techno with a low volume and a solid beat" and receive relevant results. Similar examples on Google's Github page demonstrate suitable expressions for such claims.

MusicLM's arrival comes amid the rapid growth of OpenAI chatbot ChatGPT, prompting Google to issue what The New York Times called a "fire alarm" to the tech giant in December:

To compete, the company is rolling out 20 new product updates, as well as a version of Google Search with AI-powered chatbot capabilities, the Times reported.

However, Google has said it has no plans to make MusicLM publicly available, citing a number of risks, including cultural distortion and appropriation, technological failures and, in particular, programming biases that could lead to "potential appropriation of creative content".

According to the study, found songs were found in about 1% of the sample, indicating possible copyright infringement.

"We strongly emphasize the need for further work on these risks associated with music production and do not currently plan to publish models," the study said.

The study also notes the limitations of the technique, including negation and time used in text sentences, as well as audio quality. Looking ahead, the researchers say they plan to work on "modeling higher-level song structures such as intros, verses, and choruses."

How technology is betting on an industry created by artificial intelligence