BEST TECHNICAL PAPER AWARD - AES Spring Show 2022Technological advances have enabled new approaches to broadcast audio accessibility, leveraging metadata generated in production and machine learning to improve blind source separation (BSS). This work presents two contributions to accessibility knowledge: first, a quantitative comparison of two audio accessibility methods, Narrative Importance (NI) and Dolby AC-4 BSS. Secondly, an evaluation of the audio access needs of neurodivergent audiences. The paper presents two comparative studies. The first study shows that the AC-4 BSS and NI methods are ranked consistently higher for clarity of dialogue (compared to the original mix) whilst improving, or retaining, perceived quality. A second study quantifies the effect of these methods on word recognition, quality and listening effort for a cohort including normal hearing, d/Deaf, hard of hearing and neurodivergent individuals, with NI showing a significant improvement in all metrics. Surveys of participants indicated some overlap between Neurodivergent and d/Deaf and hard of hearing participants’ access needs, with similar levels of subtitle usage in both group
http://www.aes.org/e-lib/browse.cfm?elib=21711
