Tool: MusicLM
Google recently released an AI-powered music tool called MusicLM. It is a part of their AI Test Kitchen Project, which brings research projects in the field of artificial intelligence to life. The first public release is called MusicLM and what it does is convert text prompts to music. Here is the Abstract from the project:
We introduce MusicLM, a model generating high-fidelity music from text descriptions such as "a calming violin melody backed by a distorted guitar riff". MusicLM casts the process of conditional music generation as a hierarchical sequence-to-sequence modeling task, and it generates music at 24 kHz that remains consistent over several minutes. Our experiments show that MusicLM outperforms previous systems both in audio quality and adherence to the text description. Moreover, we demonstrate that MusicLM can be conditioned on both text and a melody in that it can transform whistled and hummed melodies according to the style described in a text caption. To support future research, we publicly release MusicCaps, a dataset composed of 5.5k music-text pairs, with rich text descriptions provided by human experts.
In order to access MusicLM, you’ll need to sign up to register your interest. It took less than 24 hours to receive access, and I must say, this generative AI music composition engine is very impressive. All you need to do is enter descriptive text into the prompt window, hit Enter, and MusicLM will do it’s best to create original music that matches the description. Like ChatGPT, the better and more descriptive the prompt, the better the results. Also, MusicLM asks you to pick which music example you prefer by clicking on a Trophy icon - a perfect example of reinforcement for the AI algorithm. It gets better when you give it feedback.
For my first prompt, I entered “Create a laid back reggae groove”. Here is one of the two tracks it created:
It’s definitely reggae, and it’s definitely laid back. For now, MusicLM does not create lyrics and a voice part, but you can almost hear one quietly in the background.
For the next example, I entered “Compose music that sounds like 1920's French cafe music with an accordion and a clarinet”. This is what it created:
Pretty good - but no accordion! The music sounds muted and is pretty short - but it definitely captures the mood that I was looking for.
One thing that MusicLM doesn’t do is try to imitate an existing band. If you ask MusicLM to create a piece of music that sounds like Beyoncé, it will give you an Oops message, saying that it can’t do that.
So how can you use this with your students? I think it is a PERFECT way to have students describe the kind of music they would like to compose, enter it into the prompt window, and see and hear what MusicLM creates. Have the students give feedback - which example do they think is better - and why? Try to create a piece of music with varying levels of descriptive text. For example, ask MusicLM to “compose a piece of music that is happy” and see what you get. Next ask MusicLM to “compose a piece of music that is upbeat, with a bouncy feel”. Compare and contrast the two results. Next add more descriptive text. I think that this is a great way to approach AI with your students - especially when thinking about its impact on music.
What do you think? I urge you to register for an account and see what you can do with this powerful new AI music tool. Better yet, see what your students come up with. Let me know how it goes!