Regardless of ChatGPT, Google has recently released a MusicLM that can be compiled into various styles of music through text descriptions provided by users. Although the same type of AI is not uncommon, TechCrunch said that even Google has AudioLM, and OpenAI has Jukebox. However, the strength of MusicLM is that it can provide more diversified and in-depth music works after 280,000 hours of music database training.
MusicLM AI not only combines different genres and instruments, but can generate music according to the music concept provided by the user, making it better than other similar sound synthesis AI. For example, users want to have music that combines dance music, reggaeton (Reggaeton), space and otherworldly feelings at the same time, and make people feel puzzled and awed, and MusicLM can generate it according to this requirement. Users can even import humming, whistling, or descriptions of paintings as the basis for generating music. In addition, there is also a “story mode” to connect different descriptions together to generate works similar to DJ skewers.
Like other AIs, MusicLM will also generate some strange audio that humans cannot understand, but it is very strong for musicians, because computers can “play” music that is difficult for humans to achieve, and AI music may be as accustomed to us The arrangement of the music is different.
On the other hand, Google did not make MusicLM public, because it is not surprising that its AI works have copyright-related issues. According to the press release, about 1% of the generated work is identical to the original material in the database. Eric Sunray, currently working for the American Music Publishing Association, published a white paper in 2021, stating that the “consistency” of today’s AI music when using the original song will be enough to violate the right of reproduction. Unless the user can clearly obtain the consent of the copyright owners of all the music in the database, it will be difficult to officially publish the relevant works.