Whether you like it or not, AI is encroaching into the world of music production in a big way, and Google has now provided some of the most compelling evidence of this yet. MusicLM is a new AI model that has the ability to generate music just from text, and even at this early stage of its development, the results being produced are pretty staggering.
This isn’t the first time we’ve seen this kind of technology, but the development team at the Google Research department claims that MusicLM “outperforms previous systems both in audio quality and adherence to the text description”. We’re told that it “casts the process of conditional music generation as a hierarchical sequence-to-sequence modelling task, and it generates music at 24kHz that remains consistent over several minutes”.
As well as being able to generate “high-fidelity” music from text, you can also instruct it by humming or whistling a melody. These two approaches can be combined, opening up all sorts of creative possibilities.
To demonstrate the potential of MusicLM, Google has released dozens of audio clips alongside the text that was used to create them (opens in new tab). To give you an example, one caption reads: “A rising synth is playing an arpeggio with a lot of reverb. It is backed by pads, sub bass line and soft drums. This song is full of synth sounds creating a soothing and adventurous atmosphere. It may be playing at a festival during two songs for a buildup.”
There are also complete 5-minute tracks based on one- or two-word prompts (examples include melodic techno, swing and relaxing jazz) and demos of the Story mode, which enables you to create music based on a series of text prompts.
MusicLM can also be instructed via a combination of picture and caption, or generate audio that’s ‘played’ by a specific type of instrument, or in a certain genre. Even the experience level of the ‘musician’ can be set, and you can create music inspired by places, epochs or requirements (eg, motivational music for sports).
It’s all highly impressive and terrifying in equal measure, and will provide fuel for the fire for those who argue that AI has the potential to destroy music production as we know it. The genie is well and truly out of the bottle now, though: you can’t try MusicLM for yourself just yet, but the development team has just released a research paper and a dataset comprising 5,500 music-text pairs that offer additional insight to those who want to explore further.
Get a glimpse of the future over on the Google Research (opens in new tab) GitHub.