Google developers have showcased a new breakthrough in its artificial intelligence (AI) system designed to automatically turn text into music.
Dubbed MusicLM, the AI system was developed by Google Brain, the technology giant’s research division, and is reportedly capable of generating musical compositions in a range of styles, from classical to jazz.
The program is able to generate musical compositions that are based on input text, such as lyrics or poetry. It is also conditioned to transform whistled and humming melodies to any form of music described through a text caption.
In its Github page where the project is ongoing, the Google research team states that MusicLM is a model that generates "high-fidelity music from text descriptions such as 'a calming violin melody backed by a distorted guitar riff'".
"Our experiments show that MusicLM outperforms previous systems both in audio quality and adherence to the text description," the team added in its GitHub paper abstract.
The program is able to generate musical compositions that are based on input text, such as lyrics or poetry. It is also conditioned to transform whistled and humming melodies to any form of music described through a text caption.
In its Github page where the project is ongoing, the Google research team states that MusicLM is a model that generates "high-fidelity music from text descriptions such as 'a calming violin melody backed by a distorted guitar riff'".
"Our experiments show that MusicLM outperforms previous systems both in audio quality and adherence to the text description," the team added in its GitHub paper abstract.
The Google AI music system takes text as an input and then creates a musical composition based on the words, phrases and sentences within the text. MusicLM is based on a Generative Pre-trained Transformer-3 (GPT-3) model, which is an advanced form of the AI technology used in natural language processing (NLP).
MusicLM has been trained on a large dataset of music and its corresponding lyrics, allowing it to create music in various genres. According to Google, MusicLM can generate its own melodies, harmonies, and basslines, as well as generate accompaniments to existing melodies.
It works by analyzing the text and generating a score that is then translated into musical notes. The program is capable of generating multiple layers of musical elements, such as percussion, bass, and melody. It can also adjust the tempo and key of the composition to match the text.
The MusicLM AI system was first developed in 2011 but it is still in its early stages. Google says it could be used in a range of applications, such as creating soundtracks for films or video games, or in the music industry to generate new compositions. Although it was developed a little while ago, this new version has much more functions to boast about.
The team behind MusicLM hope that it will be able to help create unique and interesting music. The scope of the project is not yet known, but when it is finally released for public use, it will help to change the world of entertainment such as creating copyright-free tracks for content creators among other applications.
Click here to listen to an example of music generated with the MusicLM. Or visit the GitHub abstract page to listen to more of Google's successful projects with the AI.
Posts
