DNN-BASED SPEECH QUALITY ASSESSMENT FOR BINAURAL SIGNALS

Quality prediction in the scope of speech communication is still a challenging tasks even though several proprietary as well as standardized models were developed for this purpose over the past decades. Scenarios including acoustic paths were not covered adequately so far, in particular for signals that are perceived binaurally. The present work introduces a novel quality prediction approach that is based on recent deep learning techniques, but also considers important traditional aspects of psychoacoustics, auditory databases and standardized terminal testing methods.