We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
Use of synthesized higher order ambisonics impulse responses to generate virtual sound environments for hearing aids research.
- Authors
Sayin Sarac, U.
- Abstract
Real-life situations are difficult to replicate in the laboratory and, thus, often discarded during Hearing Aids (HA) optimization, leading to performance inconsistencies and user dissatisfaction. The most prominent problems for HA users, such as the known cocktail party effect or accurate sound localization, are not entirely resolved. Current solutions using ray-tracing, image source methods (IMS) and Ambisonics still have to compromise between either realism or being adaptable to cover many different scenarios. This study proposes a new toolset that incorporates real-life conditions in the design, test and fitting of HAs with the aim to produce large datasets of accurate and diverse data to use in hearing aids research, which can then be used to train more complex machine learning (ML) algorithms. This toolset includes a spatial audio simulation framework for generating a large number of IRs corresponding to realistic situations using a mixture of IMS and Ambisonics, a machine learning (ML) algorithm focused on prominent HA problems trained with the abundant newly generated data, and a low-cost spatial audio solution for audiological clinics for improved fitting of HAs. The current article presents the first results of the spatial audio simulation framework compared to a reference scenario and other existent solutions in the literature. The simulation framework is capable of representing arbitrary number of sources with their corresponding directivities within the Ambisonics domain. The resulting IRs are later on used with any hearing aids device that are previously measured for HRIR characterization through a set of measurements without the necessity of an anechoic chamber thanks to post-processing techniques. The first findings demonstrate that synthesized higher-order Ambisonics (HOA) impulse responses (IR) can be a powerful tool for generating real-life situations for HA research. The reference situation generated with synthesized IRs demonstrated a match with the original measurement in terms of signal-to-noise ratio (SNR) and SNR gain in the case of a change in directivity settings of the HA device from omnidirectional to cardioid or adaptive beamforming. The devices are characterized by a fifty-point HRIR measurement in a diffuse field environment for testing purposes. In addition to SNR and SNR benefit, the interaural time differences seem to be kept intact compared to the original measurements validating the IRs and HRIRs as suitable to use in computer generated simulations rather than actual measurements for representing different scenarios. All the results demonstrate that the synthesized HOA IRs are a possible alternative to current solutions to generate a large set of realistic data to be deployed in ML algorithm training.
- Publication
Journal of Hearing Science, 2022, Vol 12, Issue 1, p204
- ISSN
2083-389X
- Publication type
Academic Journal