Google Has an AI That Can Turn Text Prompts Into Music
If the controversy surrounding AI-generated art isn’t enough, Google has created MusicLM, a system that generates music from text descriptions, images, and simple melodies. If you think it sounds too good to be true, keep reading to find out more.
Google Has an AI Called MusicLM
While Google has yet to officially releaseMusicLM, the generative AI system is already making headlines. In its research abstract, where it samples what the technology can do, Google describes it as a model that generates “high-fidelity music from text descriptions”.
The software, which was trained using over 280,000 hours of music, provides various musical scenarios in a number of formats. This includesAudio Generation From Rich Captions,Story Mode, andPainting Caption Conditioning—a painting paired with a description of it (Google has included a sample of the AI-generated audio).
Google is not the first to utilize AI for music production. For example,LALAL.AI has a Cassiopeia algorithmusing an artificial neural network that separates the vocals and instrumental without the need for manual execution from an engineer. However, Google has taken a massive leap forward in creating a system that can generate music from text alone.
Google Won’t Release MusicLM: Here’s Why
Unfortunately for Google, it is unlikely to release MusicLM. The most obvious reason is that it infringes upon copyrighted music. When Google ran tests on it, they found that at least 1% of every song the system spat out was identical to the music it used to train the software.
That statistic might seem low, but it spells disaster, and Google isn’t ready to get wrapped up in any lawsuits.
Moreover, the technology may be unparalleled to others in its field, but there are still elements that need refining; the vocals sound buggy and distorted, and the language resembles Simlish more than any existing language in the real world.
Google Should Rethink MusicLM
MusicLM could be a big help to musicians and non-musicians alike, but unless Google makes extensive improvements, it shouldn’t go live. If the system is to be released, at least two major changes need to happen.
First, Google needs to find a way to make MusicLM more ethical.Shutterstock has made AI art generation more ethicaland accessible for users, perhaps Google could follow suit. If Google were to train MusicLM with ready-for-licensing music, as well as compensate the artists who contribute to the system, then releasing it might be feasible.
Second, Google needs to improve the vocals. If Google can’t find a way around the unrealistic vocals and gibberish, it should forgo them entirely. Software like Vocaloid would be the standard, and if it’s not meeting that same quality then the function is futile.
Not Today, Google
It’s evident that there are many complications with MusicLM, and releasing the software as is would be the wrong move. But technology advances in the blink of an eye, so there’s always the possibility that something better comes along—just not today.
Not Linux, not Windows. Something better.
Love fades, even for the best open-source darling.
My foolproof plan is to use Windows 10 until 2030, with the latest security updates.
If an AI can roast you, it can also prep you for emergencies.
The best features aren’t the ones being advertised.
You don’t need to fork out for expensive hardware to run an AI on your PC.
