Heading hidden

The first deep AI based synthesizer

We developed the first musical audio synthesizer combining the power of deep generative models and the compacity of Eurorack format

Eurorack - Deep impact synthesizer

The Neurorack is the first ever AI-based real-time synthesizer, which comes in many formats and more specifically in the Eurorack format. The current prototype relies on the Jetson Nano. The goal of this project is to design the next generation of music instrument, providing a new tool for musician while enhancing the musician's creativity. It proposes a novel approach to think and compose music. We deeply think that AI can be used to achieve this quest. The Eurorack hardware and software have been developed by our team, with equal contributions from Ninon Devis, Philippe Esling and Martin Vert.


Project Github // Youtube video

Video demonstration

Check out our demo on YouTube


More information



Deep learning models have provided extremely successful methods in most application fields, by enabling unprecedented accuracy in various tasks, including audio generation. However, the consistently overlooked downside of deep models is their massive complexity and tremendous computation cost. In the context of music creation and composition, model reduction becomes eminently important to provide these systems to users in real-time settings and on dedicated lightweight embedded hardware, which are particularly pervasive in the audio generation domain. Hence, in order to design a stand alone and real time instrument, we first need to craft an extremely lightweight model in terms of computation and memory footprint. To make this task even more easier, we relied on the Nvidia Jetson Nano which is a nanocomputer containing 128-core GPUs (graphical unit processors) and 4 CPUs. The compression problem is the core of the PhD topic of Ninon and a full description can be found [here](https://ninon-io.github.io/research/phd/).

Targets of our instrument


We designed our instrument so that it follows several aspects that we found crucial:
  • Musical: the generative model we choose is particularly interesting as it produces sounds that are impossible to synthesize without using samples.
  • Controllable: the interface was relevantly chosen, being handy and easy to manipulate.
  • Real-time: the hardware behaves as traditional instrument and is as reactive.
  • Stand alone: it is playable without any computer.

Design process

The work was divided between three major steps: ![steps](images/steps.PNG) We set our sights on the generation of impacts as it is a very complex sound to reproduce.

Model Description

We set our sights on the generation of *impacts* as they are very complex sounds to reproduce and almost impossible to tweak. Our model allows to generate a large variety of impacts, and enabling the possibility to play, craft and merge them. The sound is generated from the distribution of 7 descriptors that can be adjusted:
  • Loudness
  • Percussivity
  • Noisiness
  • Tone-like
  • Richness
  • Brightness
  • Pitch


One of the biggest advantage of our module is that it can interact with other classical synthesizer. We divided our panel into CV and gates control:


Esling, P., & Devis, N. (2020). Creativity in the era of artificial intelligence. arXiv preprint arXiv:2008.05959.

Esling, P., Devis, N., Bitton, A., Caillon, A., & Douwes, C. (2020). Diet deep generative audio models with structured lottery. arXiv preprint arXiv:2007.16170.