PhD Thesis Download
This thesis studies binaural sound reproduction from both a technical and a perceptual perspective, with the aim of improving the headphone listening experience for entertainment media audiences. A detailed review is presented of the relevant binaural technology and of the concepts and methods for evaluating perceived quality. A pilot study assesses the application of state-of-the-art binaural rendering systems to existing broadcast programmes, finding no substantial improvements in quality over conventional stereo signals. A second study gives evidence that realistic binaural simulation can be achieved without personalised acoustic calibration, showing promise for the application of binaural technology. Flexible technical apparatus is presented to allow further investigation of rendering techniques and content production processes. Two web-based studies show that appropriate combination of techniques can lead to improved experience for typical audience members, compared to stereo signals, even without personalised rendering or listener head-tracking. Recent developments in spatial audio applications are then discussed. These have made dynamic client-side binaural rendering with listener head-tracking feasible for mass audiences, but also present technical constraints. To limit distribution bandwidth and computational complexity during rendering, loudspeaker virtualisation is widely used. The effects on perceived quality of these techniques are studied in depth for the first time. A descriptive analysis experiment demonstrates that loudspeaker virtualisation during binaural rendering causes degradations to a range of perceptual characteristics and that these vary across other system conditions. A final experiment makes novel use of the check-all-that-apply method to efficiently characterise the quality of seven spatial audio representations and associated dynamic binaural rendering techniques, using single sound sources and complex dramatic scenes. The perceived quality of these different representations varies significantly across a wide range of characteristics and with programme material. These methods and findings can be used to improve the quality of current binaural technology applications.
This PhD project was carried out at the University of York's Audio Lab, under the supervision of Anthony Tew, while Chris was working at BBC R&D. The thesis is freely-available from the White Rose online repository at http://etheses.whiterose.ac.uk/24022/.
This publication is part of the Immersive and Interactive Content section