On the Assessment of High-Quality Voice Recordings including Voice Postprocessing

Evaluating the quality of voice recordings invokes two different aspects of preference: the appeal of the speaker's voice characteristics and the perceived degradation of the actual recording. Subjective experiments where no reference ideal is provided, called absolute category ratings, fuse together both components. This research investigates if voice postprocessing (timbre optimization, loudness compensations, de-essing, room reverberation, and noise suppression) can improve perceived quality. The results show that none of these processing methods significantly improve the perceived quality and mostly give rise to a quality degradation. The voice preference is a very significant factor in the perceived quality. The subjective results are used to validate ITU-T Recommendation P.863 (POLQA) in the high quality region.