HomePage

AQUA-RIUS Project: Audio Quality Analysis for Representation, Indexing and Unifying Signals

As instrumental sound ``timbre is defined as all sound characteristics which are not related to pitch, loudness and duration, we consider here ``audio quality as everything related to the sound characteristics which is not related to the content sources.

This therefore includes the choice of microphone, recording media (tapes, vinyls, digital and related potential artifacts), audio production chain (equalization, compression, reverberation) and diffusion (such as mp3 data-reduction). Some of them are considered as artifacts (or degradation), some of them are related to artistic choices.

Scientific objectives:

The project AQUA-RIUS proposes to complete an exhaustive investigation of ``audio quality''. Hence, we aim at developing innovative tools to enable a better control of the creation processes of an audio signal from the recording of the stems to the studio mastering and further up to their use in real-world diffusion contexts (such as DJ or radio mixes)~\cite{schwarz2019}. The goal is to provide analysis and synthesis tools leading to a more robust and unified data representation for audio signals.

Indeed, this research project includes fundamental research related to signal processing and efficient data representation for machine learning with a consideration to real-world application scenarios for dataset audio tagging with a possible industrial valorization.

The industrial collaborators of IRCAM and the first-rate expertise of the project contributors in audio signal processing and in machine learning (including deep learning) are a definite asset to tackle this project.

The project AQUA-RIUS will address the following scientific questions

• The analysis and modeling of audio quality with a focus on the capability to predict the effects applied during the audio signal production and diffusion chain.
• The simulation and the synthesis of audio quality effects with a consideration for making more robust machine learning algorithms through data augmentation and domain adaptation techniques to deal with several training datasets.
• The full control of the audio quality in order to cancel or to reverse production and diffusion effects.

Project details

• Scientific Project Start Date: 01/01/2023
• Project Duration: 42 months
• Project Coordinator: Dominique FOURER (Assoc. Prof. Univ. Évry, IBISC SIAM team)
• Institutional Partners: Telecom Paris (LTCI), IRCAM (UMR STMS)
• Total Funding: 510 K euros
• Funding for IBISC: 149.16 K euros

Partners and Contributors

IBISC

Telecom Paris

IRCAM