Google published new research that generates music in any genre through artificial intelligence. MusicLM produces music with impressive variety and depth with its unique model and vast training database compared to the existing AI music generators.
Chris McGrath/Getty Images
(Photo : Chris McGrath/Getty Images) ISTANBUL, TURKEY - MAY 06: A woman views historical documents and photographs displayed in a high tech art installation at Salt Galata on May 6, 2017 in Istanbul, Turkey. The "Archive Dreaming" installation by artist Refik Anadol uses artificial intelligence to visualize nearly 2 million historical Ottoman documents and photographs from the SALT Research Archive.
Google's MusicLM
As artificial intelligence progresses day by day, it is not already surprising that big tech companies are joining this new movement and adapting AI to their operations. Seeing how AI image generators became so popular in 2022, AI music generators could also see similar growth and marketability in the near future.
A study was recently published by Google on MusicLM that produces music in any genre. Engadget reported that it offers a unique model and vast training database with 280,000 hours of music, which helps the system to produce music with different variations and more profound than any AI music.
Aside from the genres and instruments, the AI also writes tracks using abstract concepts that computers and past AIs can not do. MusicLM can make melodies based on humming, whistling, or a description. Through stitched story modes, the system will be able to produce a DJ set or soundtrack.
The system could also be instructed through a combination of picture and description, in which the music will be produced through the grasped places, epochs, and requirements.
MusicLM also generates vocals, including choral harmonies. However, most of the produced lyrics were gibberish as it was sung by synthesized voices combined from several artists.
TechCrunch posted some of the sample audios that the AI has produced and reported that despite the music being far from perfect and flawless compared to real music, MusicLM is going beyond expectations with its capabilities.
Launching MusicLM
Google researchers believe that MusicLM has ethical challenges posed in its system, especially once it is released. There is a tendency for the system to incorporate copyrighted materials based on the training data that could be incorporated to the generated songs.
According to Google's released research paper, researchers found out during an experiment that 1% of the generated music by the system was directly replicating songs from its training. This is the main reason why they are discouraged from releasing MusicLM to the public.
The authors stated, "We acknowledge the risk of potential misappropriation of creative content associated to the use case. We strongly emphasize the need for more future work in tackling these risks associated to music generation."
Medium reported that Google additionally released the MusicCaps data set. This is a collection of 5500 music-text pairs that give the researchers an opportunity to have a perception regarding MusicLM's generative process.
