Google Created ‘MusicLM’, An AI That Creates Music From Text Description

Google Created ‘MusicLM’, An AI That Creates Music From Text Description

MusicLM is trained on 280,000 hours of music dataset, not going to be available publicly anytime soon

The tech giant Google has developed a powerful new AI (Artificial intelligence) system that can generate music from the text in any genre.

  • The AI can generate music in high fidelity from a text description, an audio sample, and images.
  • Due to copyright issues in the created music, Google is not making the system public before it is improved.

MusicLM isn’t the first AI system that can generate music but it is indeed the first that can generate music with high-fidelity and complex compositions. The system can create sensible songs from the description of the text. The algorithm is trained on 280,000 hours of music dataset. The system has the ability to create music on tunes that are hummed, whistled, sung, or played on an instrument.

Keunwoo Choi, an AI scientist from Gaudio Lab tweeted on Friday and praised MusicLM as better than ChatGPT. 

A Google researcher said that the system can turn music from a sequentially written description. MusicLM can be fed with a combination of images and a caption. The AI can synthesize vocals but the produced results are bad and distorted.

Google trained the AI over the songs and music that are copyrighted. Now the issue with the generated music is that it contains some portion of the music that the AI was trained on. During the experiment, the researchers found that roughly one percent of the produced music is directly copied from the songs that were used to train the algorithm.

Also Read – Youtube To Fix Policy To Demonetize Videos With Foul Language And Swearing

The figure, i.e., 1% is high enough that the tech giant has decided not to make MusicLM publicly available in its current state, noted by TechCrunch. The researchers stress the need for additional future work to address these risks associated with music production and the possibility of creative production being inappropriately used in use cases.

However, some users were amazed by the audio samples released by the tech giant.

Talking about the legal concerns, it’s not the first time. Jay-record Z’s music company filed a copyright complaint against a YouTube channel in 2020, in which AI was used to synthesize Jay-Z’s rendition of music like Billy Joel. The videos were taken down but later put back online after the request was incomplete.