Google Develops AI That Can Create Music From Text

AIs that create creative content such as text and images are on everyone’s lips these days, or rather in the eyes, but now there’s also something on the ears: Google has introduced an artificial intelligence that can create pieces of music according to the specifications of users.

For some time now, the roof has been on fire at Google and the search engine giant has triggered an internal “red alert”. The main reason for this was the success of ChatGPT, Google fears that the responses of the text or chatbot could undermine its own search engine business. The consideration or fear about this is quite simple: who needs a search engine when an AI can deliver better results for everyday questions?

In the course of the latest business results, the Mountain View, California-based company has also emphasized that in the future it will shift up at least one gear in terms of AI and fully focus on the introduction and development of related services.

MusicLM masters text-to-music

A result is now (half) there and is called MusicLM. This AI is able to create a piece of music from any text description. In a research paper on GitHub, Google presents this in more detail and of course also provides audio samples. Basically, MusicLM is described as a “model that can produce faithfully reproduced music based on textual descriptions such as ‘a soothing violin melody overlaid with a distorted guitar riff'”.

Google writes that the experiments would prove “that MusicLM performs better than previous systems in terms of both audio quality and compliance with the text description.” Google goes on to say that the AI ​​is already showing that it can be conditioned to both text and a melody, “by converting whistled and hummed melodies according to the style described in a text caption.”

TechCrunch points out that MusicLM is not the first AI capable of generating music, but no other has been able to create complex pieces in Hi-Fi quality. You can’t try MusicLM yet, but Google provides several examples along with the relevant texts. These are mostly electronic, this is where the AI ​​works best – Google undoubtedly still has to work on classic, rap and metal, these examples still sound pretty idiosyncratic to the point of being wrong.