📢 We are looking for an intern to join our music AI team this summer! PhD or MSc students or recent graduates in Generative AI/ML for Music/Audio, come work on a cool project at the world's largest music company! Details and application form below 👇 umusic.wd5.myworkdayjobs.com/UMGUK/job/King…

Proud to announce that AIMers will have a remarkable presence at IEEE ICASSP, featuring nine papers on topics ranging from graph neural networks to contrastive music representations! For details, please check out the blog post: aim.qmul.ac.uk/aim-at-icassp-…

Happy to say that our tutorial for Self-Supervised Learning in MIR was accepted ISMIR Conference 2025!! Join Alain Riou, Marco Pasini Yuexuan KONG, Gabriel Meseguer and Stefan Lattner, and myself in Daejeon in september to learn more about music SSL 🎶 Stay tuned for paper preprints 🎀

I won't be there, but the amazing Christos Plachouras 🦋 will be presenting our recent work on an evaluation framework for learned representations beyond downstream probing! arxiv.org/abs/2505.06224

🚨 Two papers at ISMIR 2025! 🔹 SLAP: Siamese Language-Audio Pretraining arxiv.org/abs/2506.17815 🔹GD-Retriever: Controllable Generative Text-Music Retrieval with Diffusion Models arxiv.org/abs/2506.17886 Thanks to my co-author Alain Riou and @ISMIR2025! More Details soon!

Who should be credited for this audio generation? In our recent work, we explore the application of unlearning methods to establish training data attribution in realistic music generative models.

Show don't tell... Magenta RealTime now supports audio prompting in addition to text prompting, so you can now use clips of audio as latent anchors to steer generation. youtu.be/vHIf2UKXmp4?si…

Tokenization has been the final barrier to truly end-to-end language models. We developed the H-Net: a hierarchical network that replaces tokenization with a dynamic chunking process directly inside the model, automatically discovering and operating over meaningful units of data

Yet another benefit of SLAP compared to CLAP: You can further decrease memory requirements using Gabriele Berton's trick! 🪄 Currently training a SLAP model with batch size 512 on a single A100 😎

Too many negative samples in our lives... 😔 If you wanna get rid of them, don't miss our 10min presentation with Julien Guinot Tuesday 23 at 2pm 👁👁 👋💥SLAP: Siamese Language-Audio Pretraining Without Negative Samples for Music Understanding Best Paper nominee at #ISMIR2025 🥳

since we're approaching the official & full release of TalkPlayData 2, i updated the demo page to show the data and the process much more clearly! talkpl.ai/p/talkplaydata…