Investigating, modelling and improving music listening for hearing impaired individual.

点击翻译按钮获取中文摘要

This project aims at improving the music listening experiences of hearing impaired individuals byexploring and understanding the limitations of current hearing aids (HAs) technologies, as well assetting the grounds for novel modelling research for better understanding the perceptualmechanisms involved, and potentially predicting outcomes from novel technologies, techniquesand tools. This will be achieved through three main work-packages (WP):WP1 - Dataset. Design of a set of labelled data for assessing music listening for hearing impairedlisteners and hearing aid users. This will include audio material (e.g. stem music tracks, listeningscenarios and acoustic environments - measured and/or simulated), and tools to carry outevaluations and produce additional data (e.g. hearing loss models and virtual hearing aids, as wellas tools for the creation of immersive an interactive experiences, such as sound spatialisers andreverb simulators). In order to maximise the potential impact, the dataset will be designedinvolving the relevant stakeholders (e.g. hearing aid users, clinicians, scientists, musicians, etc.).WP2 - Validation. Carry out an initial validation of the dataset with normal hearing participants,possibly using simulated hearing loss. An important outcome of this step is to start labelling theinput and output (e.g. perceived sound quality, etc.) data from the evaluation, and commence thesetup of an AI model to be subsequently (i.e. in the following steps of this research, or possiblybeyond this specific project) trained as artificial listener/predictor.WP3 - Explorations. By using the developed dataset and AI model, further explore differentaspects of the research. The choice on the focus of this stage will be made mid-project, but specificdevelopments might include:- Interface/control. Develop a prototype of personalisable/scalable interface for controllingthe various parameters of the MP. This could be based on AI, understanding from the userwhat can/should be controlled and how, and ultimately producing a prototype of theirown customised control interface.- Optimal acoustics. Optimising the acoustics of the spaces (either real or virtual) for aspecific listener. It might be interesting here to compare computational optimisation3methods with behavioural assessments, aiming again at further training the AI model, andpossibly developing a prototype optimiser for further evaluations beyond this project.- Optimal signal processing choices. As above, but related specifically to the calibration ofthe MP.