진학프로석·박사 채용 접수 플랫폼

The position offered is a five-year post-doctorate as part of the ERC Advanced Grant G3S project.

Presentation of the G3S project

Machine learning is particularly well suited to dealing with ill-defined problems, where complexity precludes direct solutions; it is therefore of vital importance in the field of music creation, where compositional knowledge is far from explicit. Until now, the generation of sound spaces has often been neglected in the application of AI to music and sound, whereas 3D spatial audio is the subject of major industrial developments and standardisation.

The ERC Advanced Grant project G3S (Generative Spatial Synthesis of Sound and Music), designed by and for musicians and sound creators, will explore generative approaches to spatialization based on machine learning, to break away from the standards that format our ways of creating or perceiving spatiality, by opening up to all the ways and processes of spatial audio. It is positioned in an original way at the rarely explored intersection of spatialisation and AI, the latter considered in a frugal, local and open source way, and within the framework of an international collaborative research and creation network.

From a large set of varied musical pieces proposing a construction of the sound space, from which we will recover the sound engines (represented by the operations on the signal), their multichannel recordings and the semantic descriptions of spatiality, we will train low-dimensional learning models, combining existing neural techniques. These models will be used to generate sound spaces and to explore them by means of user requests, either functional (describing the desired processing) or imitating an audio result, or semantic (describing the desired space). The spatial engines selected by the user can be exported in the form of plugins.

Over five years, the G3S project will implement and articulate four major research objectives: 1) the design of a unified operational representation of existing spatialization engines 2) the proposal of a thesaurus and quantitative measures to describe the spatiality of sound 3) the generation of sound spaces from machine learning 4) the design of user interfaces to explore spatial audio. We will produce open source environments, compatible with audio standards and shared with the computer music community; we will validate them by commissioning composers, creating works, workshops and concerts.

Details of assignments

As part of the project team under the scientific responsibility of P.I. Prof. Alain Bonardi, you will be associated for five years with two other post-docs to carry out original, high-level research, exploring generative approaches to sound spatialisation based on machine learning, largely founded on research-creation methods. The team will also include four PhD students (each working on a subject corresponding to one of the four research objectives mentioned above).

The general duties of your post will involve contributing to :

• research and research-creation carried out as part of the G3S project;
• setting up and running the international collaborative research and creation network with which the G3S project will be in regular contact;
• producing project deliverables, including publications and communications;
• organising and running major scientific, musical and dissemination events to help evaluate and disseminate the results of the project;

This post-doctorate opportunity is focused on the theme of 'Sound spaces/spatialisation in the context of electroacoustic and mixed music', which is one of the three areas of the G3S project (the other two being 'Interactions and interfaces' and 'Machine learning and generativity'). The 'Sound Spaces' domain will be explored from multiple angles: from the construction of sound engines (operations on signals at different levels of representation) to the qualitative or quantitative description of spatiality (use of existing descriptors or creation of new descriptors, either on output signals or from microphone recordings).

Throughout the project, the research work on sound spaces will be closely linked to the work on machine learning and interfaces for exploring generated proposals: you will explore a generic framework for modelling operations on signals, opening up the possibility of manipulations and reformulations by generative processes, as well as a set of descriptions of spatiality.

ACTIVITIES

Over the five years of the project, you will be able to consolidate your experience as a young researcher, contribute to publications in the most reputed conferences and journals in the fields covered, grow your international research network (thanks in particular to the project's collaborative network) and meet industrial R&D players.

The general activities of your post-doctorate will be in line with the general missions defined above:

• conducting your post-doctoral research and research-creation, in collaboration with other post-doctoral and doctoral students, under the scientific responsibility of the P.I. of the ERC G3S project. You will take part in the development and musical and sound testing of machine learning-based prototypes for exploring spatiality;
• work on the various project deliverables:
  • production of open source software
  • production of high-level international publications;
• the organisation of workshops for a variety of audiences, study days, leading symposia on the themes of the project, and professional or amateur concerts using the results of the project.

The activities specific to your post-doctoral research will combine theoretical and practical work relating to the exploration of spatiality, in relation with professional artists and, more broadly, user communities. Working more particularly with the two PhD students who will be working on generic operational modelling and the description of sound spaces, you will be required to take part in the collection of pieces with electronics (from the international collaborative network) contributing to the learning of our models, to constitute a significant state of the art in the field, and to build models and prototypes that will be evaluated in our 3D dome and in concert situations.