Audio Source Separation Recent Advancement, Applications, and Evaluation

Product Development

Chair:
Chungeun Kim, University of Surrey - Guildford, Surrey, UK; Sony Interactive Entertainment Europe - London, UK
Panelists:
Jon Francombe, BBC Research and Development - Salford, UK
Nima Mesgarani, Columbia University - New York, NY, USA
Bryan Pardo, Northwestern University - Evanston, IL, USA

Audio source separation is one of the signal processing techniques inspired by the humans' corresponding cognitive ability in auditory scene analysis. It has a wide range of applications including speech enhancement, sound event detection, and repurposing. Although initially it used to be only possible to separate the sources in very specific capturing configurations, only to a suboptimal level of quality, advanced signal processing techniques, particularly deep learning-related approaches, have both widened the applicability and enhanced the performance of source separation. In this workshop the recent advancement of source separation techniques in various use-cases will be discussed, along with the challenges that the research community is currently facing. Also, research activities on the quality aspects specific to source separation towards effective performance evaluation will be introduced.