Google Brings DALL-E Technology In AI Baru That Can Product Music From Text

Illustration (Photo myselfwar Bachu / Unsplash)

JAKARTA - Only with text, Google's new artificial intelligence (AI) is capable of producing music that lasts a few minutes. Called MusicLM, the way it works is similar to OpenAI's DALL-E's startup.

This paper includes a large number of samples made using MusicLM, such as five-minute clips from techno, swing, jazz melodies, various genres from meditation, electronic music to death metal and rap.

In addition to producing musical works that last a few minutes from a text order, MusicLM can turn the whistle melody into another instrument.

He can produce opera vocals with the "Bellamen's" melody sung. Others, MusicLM is also able to produce songs from the gym prompt that have incoherence and vocal lyrics and melodies with Arabic-pop distinctive sounds.

Seeing this, the workings of the MusicLM system are very similar to the DALL-E which produces images from written orders. Unfortunately, at this time Google has not allowed anyone to try the music-producing AI tool themselves.

And Google also doesn't have a plan to release a model at this point. This is because researchers are worried about the potential risk of abuse of creative content and the potential for cultural abuse or misrepresentation, aka plagiarism.

Instead, Google has uploaded many of the samples produced using the model. There are 30-second footage that sounds like actual songs but it turns out to be made from descriptions along the paragraph.

This is in order to determine the genre, vibration, and even certain instruments, as well as the five-minute work produced from one or two words such as melodic techno.

Finally, quoted from The Verge, Monday, January 30, Google stated that it openly released a data collection with about 5,500 music-text pairs, which could assist in training and evaluating other music-based AIs.