I have spent quite a number of hours with MusicLM in AI test kitchen. To me, that is actually a new form of music synthesis. Synthesis in the context of FM, subtractive, etc.
I suspect MusicLM had some tracks removed from the training data because it use to make absolutely wildly creative psytrance clips and now it just doesn't.
Suno on the other hand to me has basically been a complete joke. The training data is just not wide enough to do anything interesting and new.
I think the only way to really use these tools creatively would be to produce your own trained model with AudioLM. The creative use then is in the musical output and not selling the model as yet another SaaS for $x.99 a month.
Of course, you could use Suno to create loops and sample material. The way midjourney is the greatest thing possible for digital collage in Krita/Photoshop.
GenAI art though on its own be it midjourney, stable diffusion, Suno, MusicLM, whatever already jumped the shark for me a year ago. The problem ultimately is the output is just so very limited.