The Sebass-DB: A Consolidated Public Data Base of Listening Test Results for Perceptual Evaluation of BSS Quality Measures
Thorsten Kastner, Jürgen Herre · 2022
For the development of new and improvement of existing perceptual quality measurement methods for Blind Audio Source Separation (BASS), a large and comprehensive body of subjective reference ratings is necessary, originating from well-conducted listening tests, that enable calibration and testing. To support the community in the advancement of perception-based quality measures for BASS, we published the SEBASS-DB, i.e. the Subjective Evaluation of Blind Audio Source Separation Data Base. The SEBASS-DB is a meta dataset containing the results from five high-quality MUSHRA based listening tests on the Basic Audio Quality of (blindly) separated audio source signals and can be freely used for non-commercial purposes. It contains a total of over 11,000 ratings from over 900 rated audio signals submitted to the audio source separation campaigns 2007, 2008 and 2018 as well as listening test results from three other publications by the authors. Major parts of the listening tests were conducted by expert listeners.