Virtual acoustic environments for the evaluation of hearing devices

Listening tests are essential for optimising and evaluating novel signal processing concepts for hearing devices. When they are performed inside the laboratory they are highly controlled, but typically lack ecological validity. On the other hand, field-studies provide high ecological validity, but are difficult to control, time-consuming and costly. Even though field-studies may ultimately be required, there still is a necessity for more realistic laboratory-based listening tests. This thesis focuses on creating and validating realistic virtual sound environments (VSEs), primarily based on the method of higher-order Ambisonics (HOA), for testing hearing aids (HAs). Particular emphasis has been given on how the VSE limitations affect the output of multi-microphone directional HAs. To systematically and objectively analyse the errors incurred by such VSEs, we developed and verified a framework that consists of: (1) the simulation of a reverberant acoustic scene; (2) the coding of that scene using HOA; (3) the reconstruction of the VSE via a loudspeaker array; and (4) the evaluation of the SNR benefit and output pressure of directional HAs worn on an acoustic manikin placed inside the VSE. A listening experiment was additionally designed to test the simulation results. For that purpose, a real-room 'cocktail-party' acoustic scene was created and acoustically modelled. Ambisonics was then applied to reproduce that scene inside a loudspeaker array. Listening tests involving hearing-impaired subjects, fitted with directional HAs, were next conducted inside the real and the virtual sound environments. The intelligibility and acceptable-noise-level benefits of the directional HAs were thereby estimated. The above evaluation illustrated that most of the real-environment properties can be faithfully reconstructed by the considered VSEs. Moreover, directional HA algorithms were shown to behave and adapt in a similar way as in real environments. However, sound-field reconstruction deviations were noticeable. For the case of highly directional schemes, such as the considered bilateral beamformer, such deviations may result in a reduced benefit estimation. Hence, we can conclude that although the described VSEs illustrated a reduced sensitivity in demonstrating real-life benefits, they are nonetheless suitable for evaluating directional HA processing schemes.