R&D DSP internship

Job description

In charge of Arturia's digital product audio engines (synthesizers and effects), the "Digital Signal Processing" (DSP) team relies on a wide range of techniques, such as:

  • Electronic circuit simulation and virtual analog
  • Black-box modeling
  • Physical modeling
  • Pure algorithmic approaches


As part of this team, you will be working closely with the "Product Management", "Sound Design" and "Software Development" teams and integrate our R&D processes, from developing Python/Matlab prototypes to implementing them in our public products.


Find below the two intership offers.


Digital Signal Processing / Pitch Shifting and Audio Analysis 
5 to 6 months internship


The internship consists in exploring different methodologies for pitch shifting of audio signals, with a particular focus on methods suitable for vocal chain effects.


_Subject of the internship:
You will be tasked with studying the state of the art of pitch shifting methods, to gain an understanding of the different problems encountered when pitch shifting of audio signals, and specifically the methodologies that exists to allow a natural sounding pitch shift of a singing voice.
Based on the knowledge from this preliminary work, you will then focus on prototyping different algorithms.
Another aspect of the internship will be the evaluation of the performance of the different methodologies. Through audio analysis, the performance of the implemented prototypes, as well as existing products, should be compared, using appropriate audio descriptors to define an objective comparison method.


_Profile – What we expect:
• Enrolled in a master’s degree or Engineering school (final year)
• Strong knowledge in general DSP. Some experience with audio analysis would come in handy.
• Experience in Python/Matlab programming. Some experience in C++ would be a plus for a real-time implementation.
• Good level of English, as the team is international and communication with the internship tutor will be in English.



Audio DSP – Fitting non-linear systems using machine learning algorithms

6-month internship

One of the main challenges in emulating analog synthesizers is to model each one of its modules (filters for instance) as accurately as possible. This includes all of their non-ideal behaviors (non-linearities, artifacts, etc.) which have a serious impact on the instrument timbre or grain. For this, we can rely on numerous tools and methods from our toolbox.


However, several adapted methods may give slightly different results for a same module: it then becomes difficult to evaluate which method (and with which parameters) best fits the hardware unit’s behavior.


The objective of this internship is to develop a Python tool that would solve this issue. When modeling a given module, the algorithm would compare several methods selected from a dictionary and sort them according to how well they fit the behavior of the module. The core of the internship would thus be to explore different machine learning algorithms to optimize the parameters of these models, while exploring different cost functions/metrics to compare and sort them. Different cost functions would notably allow us to focus on specific aspects of the audio signals at hands, by for instance comparing their spectrum instead of comparing them in the time domain.


_ The goals for Arturia are:
• Creating a dictionary/set of methods for modeling non-linear audio systems from the methods in our toolbox
• Exploring and creating a set of relevant cost functions for comparing methods
• Developing a Python tool able to compare methods from the dictionary on some input/output measurements and return the optimal one together with its parameters
• Modeling of a Ladder filter (4-pole low-pass resonant filter) as a first application of the algorithm


_Profile – What we expect:
• A good knowledge in Digital Signal Processing (time/frequency analysis, transformation of signals) and in Machine Learning (optimization methods) is necessary.
• It is also important to have some experience in Python and its frameworks for machine learning such as Tensorflow and/or Keras.
• Some experience in C++ Programming would be a plus.



Job requirements


Location : Montbonnot-Saint-Martin, France.


Please note that only applications from students registered in EU universities/schools will be considered, and that no remote work is foreseen.


If you are willing to join passionate teams and a growing international company in the music industry, send us your resume and cover letter, specifying the internship for which you apply.