Douglas Eck
Last reviewed
Jun 8, 2026
Sources
12 citations
Review status
Source-backed
Revision
v1 · 1,604 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 8, 2026
Sources
12 citations
Review status
Source-backed
Revision
v1 · 1,604 words
Add missing citations, update stale details, or suggest a clearer explanation.
Douglas Eck is an American computer scientist and machine-learning researcher known for his work on generative models for music and other creative media. He is a Senior Research Director at Google DeepMind, where he leads research on Generative Media, the organization responsible for Google's models that generate images, video, 3D content, music, and audio. He is best known as the founder of Magenta, an open-source research project he started in 2016 to explore how machine learning can support, rather than replace, human creativity in music and art.[1][2]
Eck's career has moved across cognitive science, neural networks, and large-scale music systems. As a postdoctoral researcher in the early 2000s he was among the first people anywhere to apply Long Short-Term Memory (LSTM) recurrent networks to music composition. He later helped build the recommendation systems behind Google Play Music before moving into Google Brain and founding Magenta. By 2025 the research carried out under his leadership underpinned widely used generative-media systems, including the Lyria music models, the Veo video models, and the Imagen image models.[1][9][11]
Eck earned a PhD in computer science and cognitive science from Indiana University in 2000.[1][3] His doctoral research sat at the boundary between artificial intelligence and the cognitive science of music. It focused on how the perception and production of rhythm and metrical structure can be modeled with networks of coupled relaxation oscillators, a simplified mathematical description of the firing dynamics of neurons. In a series of papers from this period, including collaborations with Indiana colleagues Michael Gasser and Robert Port, he showed how such oscillator networks could both perceive and reproduce the pulse and meter of rhythmic patterns, and could even learn to prefer a meter to which they had previously been exposed.[4] This early grounding in the temporal structure of music, and in biologically inspired models of timing, would shape much of his later research.
After completing his doctorate, Eck took a postdoctoral fellowship at the Dalle Molle Institute for Artificial Intelligence (IDSIA) near Lugano, Switzerland, where he worked from 2000 to 2003 under Jurgen Schmidhuber.[1][3] IDSIA was then one of the very few places developing LSTM, the recurrent-network architecture introduced by Sepp Hochreiter and Schmidhuber. Eck became one of the first researchers to apply it to a creative domain. He later recalled that "there was a point in time where there were three of us in a room in Manno, Switzerland, who are the only people in the world using LSTM."[2] With Schmidhuber he published "A First Look at Music Composition using LSTM Recurrent Neural Networks" in 2002, a proof of concept showing that an LSTM network could learn the long-term structure of a musical form, in this case the blues, and improvise new melodies over it without losing the underlying chord progression.[5] The result anticipated, by more than a decade, the sequence-modeling approaches that would later dominate music generation.
In 2003 Eck joined the University of Montreal as a faculty member in its machine-learning group, the laboratory now known as Mila and associated with Yoshua Bengio.[1][3] He rose to the rank of associate professor in computer science. His research there moved toward applied music intelligence: expressive music performance, automatic tagging and classification of large music-audio collections, and music recommendation. This combination of deep learning, audio, and recommendation foreshadowed the role that first brought him to industry.
Eck joined Google in 2010. His initial work was in music recommendation, and before founding Magenta he led the search and recommendation team for Google Play Music, applying machine learning and collaborative filtering at the scale of a commercial streaming catalog.[1][2] He subsequently moved into Google Brain, the company's deep-learning research organization, where his interests returned to generation rather than retrieval.
In 2016 Eck founded Magenta, describing it as a project "exploring the role of machine learning in the process of creating art and music."[1] The effort was previewed at the Moogfest music and technology festival in May 2016 and launched publicly on June 1, 2016, built on top of TensorFlow.[6] From the start Eck framed Magenta around augmenting human artists rather than automating them, arguing that the goal was to "build interesting ways to make new kinds of art" that give musicians and visual artists new instruments to play with.[2] Magenta released a steady stream of open-source models, datasets, and tools, and it became one of the most prominent public efforts at the intersection of AI and creativity. Notable outputs included NSynth, a neural audio-synthesis model paired with a dataset of more than 300,000 musical notes drawn from roughly 1,000 instruments, used to blend and interpolate between timbres; the Music Transformer, a 2018 attention-based model that used relative attention to generate piano performances coherent over minutes rather than seconds, which Eck co-authored; and Magenta Studio, a 2019 suite of tools that packaged Magenta's generative models as plug-ins for the Ableton Live music workstation.[7][8]
When Google Brain and DeepMind merged into a single organization, Google DeepMind, in 2023, Eck's generative-media research became part of the combined lab. He was named a Senior Research Director.[1]
As Senior Research Director for Generative Media, Eck leads and co-leads the teams behind Google's flagship creative-generation models, which span image, video, 3D, music, and audio.[1] His organization's text-to-music work traces a line from Magenta through MusicLM, a 2023 model that generated several minutes of coherent audio from natural-language descriptions, to the Lyria family of music models. By 2025 these had progressed to Lyria 3 and a real-time variant, Lyria RealTime, exposed through a public API for interactive music creation.[9][12] On the visual side, the same generative-media organization produces Imagen for image generation and Veo for video generation, both of which Eck has helped shepherd through research and into Google products.[9][10][11]
Eck has consistently emphasized tools that put generative models directly in the hands of creators. His group built the Music AI Sandbox, an experimental set of tools developed with musicians and producers, and shipped consumer-facing features such as MusicFX. At Google I/O 2025 the company presented an expanded slate of generative-media models and tools from this organization, including new versions of Veo, Imagen, and Lyria along with broader access to the Music AI Sandbox.[11] Across these projects Eck has framed the central research question as human-computer interaction as much as modeling: how to design systems that musicians, filmmakers, and designers actually want to use, and that extend rather than replace their craft.[1] He has also championed Google's People + AI Research (PAIR) initiative, which studies the human side of artificial intelligence through research, tools, and design frameworks.[1]
| Year | Project | Eck's role and contribution |
|---|---|---|
| 2002 | LSTM music composition | Co-authored, with Jurgen Schmidhuber at IDSIA, an early demonstration of LSTM networks generating blues music with long-term structure[5] |
| 2016 | Magenta | Founder of the open-source research project on machine learning for music and art[1][6] |
| 2017 | NSynth | Neural audio synthesis model and large note dataset for generating and blending instrument timbres[7] |
| 2018 | Music Transformer | Co-author of an attention-based model generating piano music coherent over minutes via relative attention[7] |
| 2019 | Magenta Studio | Suite of Ableton Live plug-ins putting Magenta's generative models in musicians' hands[8] |
| 2023 | MusicLM | Text-to-music model from his generative-media organization[12] |
| 2023 to 2025 | Lyria, Veo, Imagen | Generative-media models for music, video, and images developed under his leadership[9][10][11] |
Eck is widely regarded as a pioneer of applying deep learning to creative work. His early LSTM music experiments are frequently cited as foundational to the modern field of neural music generation, and Magenta helped popularize the idea that machine learning could be a creative collaborator rather than merely an analytical tool.[2][5] His research is heavily cited across machine learning, music information retrieval, and human-computer interaction, and he is a regular keynote and invited speaker at venues spanning technology and music, including a Centre for Interdisciplinary Research in Music Media and Technology (CIRMMT) Distinguished Lecture on music recommendation and discovery at scale.[3]
Beyond individual papers, Eck's influence is organizational: he built and grew a research program that turned experimental generative models into products used by millions, and he has been an advocate for releasing creative-AI research openly so that artists and developers outside Google can build on it. As of 2026 he continues to serve as a Senior Research Director at Google DeepMind, leading its Generative Media research.[1][9]