With the rise of the popularity of AI, we can observe new tools being created and published almost every day. From chatbots to image generators or even bots that can create music based on your prompt. And this is the case with MusicLM which Google researchers recently introduced. It’s an AI that can generate up to 5 minutes long musical pieces based on your text or can even transform your humming or whistling into other instruments. Unfortunately, it is not possible to play with this tool yet, but the creators provided us with some examples of AI in action.
The producers of MusicLM claim that their model “outperforms previous systems both in audio quality and adherence to the text description” The examples are pretty crazy, on the official website we can find some snippets of songs that were generated by long descriptions, such as “The main soundtrack of an arcade game. It is fast-paced and upbeat, with a catchy electric guitar riff. The music is repetitive and easy to remember, but with unexpected sounds, like cymbal crashes or drum rolls.”. In this case, we can listen to 30 seconds of a song that is exactly the same as its prompt! There are also 3 examples of whole 5 minute songs, such as melodic techno or relaxing jazz which are pretty convincing. I definitely can say that relaxing jazz was helpful during the writing of this blog.
I think that one of the more interesting ways to use this tool is the “Story Mode”. Here we can see one or two minutes examples of songs that were generated based on the sequence of prompts. For example, a person asked the system for an “electronic song playing in the video game” for the first 15 seconds and then for a “meditation song played next to a river” for another 15 seconds. The song perfectly blended between these two, making the change almost unnoticeable and making the song “flow” between different prompts. Then, there is the feature of changing the melody to other instruments. Here we can listen to Bella Ciao being hummed and then, the same melody was recreated by other instruments, such as guitar or saxophone. We can’t omit the ability to create music based on the place and time it would play, such as prison escape or 80s club. The tool is capable of generating vocals as well, but the quality of them is.. questionable. MusicML is able to get the tone right, but the “vocal” is just gibberish so far.
“We strongly emphasize the need for more future work in tackling these risks associated with music generation — we have no plans to release models at this point,” Google said in the research. They are planning to work on lyrics generation and improvement of vocals and more complicated song structure as well, according to the research paper.
I would say it’s a very interesting project that will attract many people. I, myself can’t wait to use it and try it out. I will definitely follow the development of this AI tool and strongly encourage everyone to check this out.
Sources:
https://google-research.github.io/seanet/musiclm/examples/
https://www.theverge.com/2023/1/28/23574573/google-musiclm-text-to-music-ai
MusicLM is fantastic! The samples are quite persuasive, and I’m looking forward to the “Story Mode” option. I can’t wait to give it a try. Although Google has some copyright issues, I am confident these will be resolved. Following the evolution and anticipating what comes next. Bravo to Google!
Wow! That’s so amazing that we have access to such tools that we couldn’t dream about even 5 years ago. This tool opens so many business possibilities, even for those who are not connected with music. As we create more immersive internet experiences, perhaps in the future websites will play music when you open them to enclose the experience. For instance, this kind of calming music that we have in the elevators. I look forward to seeing what will happen in the future with this technology, especially taking into consideration the copyright issue.
It is truly unbeliveable how big advancements we have made throughout the past few years in this area. Although I completely understand creators’ reasoning behind not releasing the product.
Thats interesting I’m really curious what are the things that AI can bring us in the future.
Very interesting. It arises the same ethical and copyright questions as other AI generators, nevertheless I see a bright future here. As much as I do not welcome it for music industry in general and can already imagine controversies incomming, some industries, like indie games will truly welcome this innovation with how cost efficient it will be.