Conversational music generation: Intuitive tools and spaces for creative flow

Digital audio workstations rely on explicit control over musical parameters: waveforms, MIDI notes, track timelines and effects. Generative AI introduces a complementary top-down approach, bridging the gap between high-level creative intuition and granular audio manipulation. This presentation explores AI-assisted music generation and interaction through the lens of user experience and interactive workflows. Focusing on the Lyria 3 ecosystem’s capabilities, we will deep-dive into the workflows available in Flow Music and show how modern generative systems can act as flexible, intuitive tools for creators. Through live demonstrations of full-song construction, we will highlight how conversational producer agents support user agency, and showcase interactive “spaces” that allow creators to develop custom tools and experiment rapidly with complex musical concepts.

Dr. Cătălina Cangea is a Staff Research Scientist at Google DeepMind, where she focuses on creating multimodal, music-centric and artist-first tools. She has previously co-led DeepMind’s GenMusic team and other music generation projects. Her passion for supporting creativity has always been fueled by her background as a musician and poet. She led the post-training and evaluation workstreams for Lyria 3, unlocking state-of-the-art music generation, and has recently been developing creative capabilities in Veo and Omni. A Cambridge alumna (BA & MPhil in Computer Science, PhD in Machine Learning), Cătălina has over a decade of AI experience heavily geared towards multimodal learning. In this showcase, she returns to Cambridge to present live demos of Lyria 3 and Flow Music, exploring new frontiers in precision and creative agency within music AI tools.