Job Description
_Internship description:
Recent research work investigated deep learning methods for modeling audio effects, that is employing neural networks to emulate the behavior of audio hardware.
In particular, a specific architecture was designed and proved to achieve good audio quality on the modeling of dynamic range compressors. The trained models are implemented in Python, with PyTorch, and leverage GPUs to apply a large number of operations per output audio sample. Therefore, these models are not ready to be running in real-time on standard consumer hardware.
Thus, the main objective of this internship is to adapt trained models into more efficient implementations, enabling real-time use on CPUs, and thereby embedding them in audio plug-ins.
_Internship objectives:
- Implement the SPTMod architecture in C++ using the Arturia framework.
- Conduct a bibliography on model compression, exploring techniques such as pruning, knowledge distillation, and neural architecture search.
- Apply these techniques to concrete cases of effect modeling and compare them.
- Evaluate and optimize both the audio fidelity and the real-time viability of modeling.
- Optimize the final C++ implementation.
Requirements
_Profile:
- Preferably PhD student, or master-level or final-year engineering school student.
- Solid knowledge in Digital Signal Processing (DSP) and Machine Learning (ML).
- Proficiency in Python and previous experience with ML frameworks (notably PyTorch).
Strong interest in research. Good command of English.
- Experience with C++ programming, audio DSP, good mathematical reasoning and communication skills are a strong plus.
_Conditions:
6-month internship in the DSP-Machine Learning team, within our R&D department.
_Location: Grenoble (Montbonnot-Saint-Martin), accessible by public transportation.
Does this opportunity sound like a good fit for you? Don't hesitate to apply: we value diversity in our workforce and are committed to equal opportunity and inclusion. This position is open to everyone.