Google has trained an “experimental AI” to generate high-resolution songs from text prompts. Now it is open to the public

In January, Google introduced MusicLM, an “experimental AI” tool that can generate high-fidelity music from text prompts and hums.

The tool is Now available for the public to try.

Google explains that at the public use level, the tool works by typing in a prompt like “Soulful jazz for a dinner party.”

The MusicLM model then creates two versions of the requested song for the person entering the prompt. You can then vote on which one you prefer, which Google says will “help improve the AI ​​model”.

The model was trained on five million audio clips, which corresponds to 280,000 hours of music at 24 kHz.

At the time of its January launch, Google released a set of examples of the From Rich Captions tool’s audio generation capabilities, the results of which you can view listen here.

Google claims, “Whether you’re a professional musician or just starting out, MusicLM is an experimental tool that can help you express your creativity.”

The company released a “look behind the scenes used yesterday at MusicLM by a sound artist, a Google Arts & Culture Artist in Residence and a Google researcher:

Google also released a paper in January detailing the research that went into developing the tool.

According to the Google researchers, “future work could focus on generating lyrics and improving lyric conditioning and vocal quality.” Another aspect is modeling high-level song structures such as introduction, verse and chorus.”

The research paper, which suggests that MusicLM “further expands the suite of tools that assist people in creative music tasks,” also added that “several risks are associated with our model and the use case it addresses.”

According to the researchers, one of the risks is that the “samples generated will reflect the biases present in the training data, raising questions about the suitability of music generation for cultures that are underrepresented in the training data, while also raising cultural concerns.” “Appropriation”.

Another risk highlighted in the paper was the “possible misappropriation of creative content”.

The researchers explained: “In line with responsible model development practices, we conducted an in-depth study of memorizing, adapting and extending a methodology used in the context of text-based LLMs, focusing on the semantic modeling phase.”

“We strongly emphasize the need for further work to address these risks associated with music creation – we have no plans to release any models at this time.”

Google MusicLM research paper

They said they “found that only a tiny fraction of the examples were memorized accurately, while we were able to find approximate agreement in 1% of the examples.”

And then added: “We strongly emphasize the need for further work to address these risks associated with music creation – we have no plans to release any models at this time.”

“Seven years into our journey as an AI-first company, we are at an exciting inflection point.”

Sundar Pichai, Google and Alphabet

Google’s surprise public release of MusicLM this week comes on the same day that Google and Alphabet CEO Sundar Pichai announced a massive push into AI with a series of AI-powered updates to various Google products.

“Seven years into our journey as an AI-first company, we are at an exciting inflection point,” Pichai said in his keynote address at the Google I/O 2023 event on Wednesday (May 10).

“We have an opportunity to make AI even more helpful to people, businesses, communities and everyone.”

The company is part of Google’s new AI push expand Its conversational AI tool and chat GPT competitor Bard is available in over 180 countries after an initial launch in the UK and US.

Bard was also recently transitioned to its “modern language model” by Google. palm tree 2. Google says this is “a far more powerful large language model that has “advanced math and reasoning capabilities.” coding functions“.

The release of MusicLM comes at a time of growing unease about the use of generative AI in music.

One of the main reasons the industry is concerned about using generative AI trained on other music is the risk of copyright infringement.

Last month, AI-generated music productions mimicking the vocals of superstar artists dominated headlines after a song titled ” heart on my sleevefeaturing AI-generated vocals copying the voices of Drake and The Weeknd went viral.

Uploaded by an artist called Ghostwriter, the track was subsequently deleted from platforms such as youtube, Spotify and other platforms. A confirmation appeared on YouTube on Ghostwriter’s Holdings page as to what triggered the title’s removal from that platform now discontinued YouTube upload.

It read: “This video is no longer available due to a copyright claim by Universal Music Group.”

Speaking of Universal Music Group‘S Results announcement for the 1st quarter Last month, Sir Lucian GraingeCEO and Chairman of Universal Music Group, remarked, “Unlike its predecessors, much of the latest is generative AI.” [i.e. ‘fake Drake’] is schooled on copyrighted material, which clearly violates the rights of artists and labels and will put platforms totally at odds with the partnerships we have with our artists and those driving success.”

In his opening remarks to analysts on the same conference call, he said: Sir Lucian Grainge also criticized the “content oversupply” that is currently emerging 100,000 titles distributed to music streaming services per day.

“Not many people realize that AI has already been a major contributor to this content glut,” Grainge said.

“Most of this AI content on DSPs is from the earlier AI generation, a technology that is not trained on copyrighted intellectual property and produces very poor quality output that is practically unappealing to the consumer.”

The rise of AI platforms that allow users to create large amounts of tracks at the click of a button has also revealed the potential for the use of generative AI streaming scam.

Earlier this month, AI-powered music creation app Boomy was unveiled, with users creating it 14.4 million Songs to this day said that Spotify had switch off its ability to upload songs to the DSP and that some already uploaded tracks have been removed.

A Spotify spokesperson later confirmed to MBW that those “certain catalog releases” were removed from Boomy because the streaming platform had detected artificial streaming of those tracks. (There was no indication that Boomy himself was involved in artificial streaming).

Boomy said on Saturday (May 6): “Curated delivery to Spotify New releases from Boomy artists has been reactivated” the company wrote on its Discord server on Saturday (May 6).

While Spotify confirmed some tracks were no longer available, it turns out they were likely Boomy’s own distributors – Downtown-own DashGo – which had stopped uploads to Spotify.

Only a small portion of Boomy titles appeared to be “greyed out” so that they could not be played. As of Monday (May 8), there were no grayed-out tracks in Boomy’s playlists on Spotify.music business worldwide