Arturia offers a 6-month DSP / Machine Learning internship (Montbonnot-Saint-Martin, near Grenoble) turning neural-network audio-effect models into real-time C++ implementations for audio plug-ins.

Internship description

Recent research investigated deep learning methods for modelling audio effects, i.e. neural networks emulating audio hardware. A specific architecture proved to achieve good audio quality on modelling dynamic range compressors. The trained models are implemented in Python with PyTorch and leverage GPUs, so they are not ready for real-time use on standard consumer hardware. The main objective is to adapt trained models into more efficient implementations enabling real-time use on CPUs, and embedding them in audio plug-ins.

Internship objectives

Implement the SPTMod architecture in C++ using the Arturia framework.
Conduct a bibliography on model compression (pruning, knowledge distillation, neural architecture search).
Apply these techniques to concrete cases of effect modelling and compare them.
Evaluate and optimize audio fidelity and real-time viability.
Optimize the final C++ implementation.

Profile required

Preferably PhD student, or master-level / final-year engineering school student.
Solid knowledge of Digital Signal Processing and Machine Learning; Python and ML frameworks (notably PyTorch).
Strong interest in research; good command of English.
A plus: C++ programming, audio DSP, good mathematical reasoning and communication skills.

Conditions

6-month internship in the DSP-Machine Learning team, R&D department · Grenoble (Montbonnot-Saint-Martin), accessible by public transportation.

DSP Machine Learning Intern