Speaker
Description
Room impulse responses (RIRs) are used to characterize rooms, their acoustic behavior and the resulting coloration of the sound. They are widely used in applications for room acoustic design, but also in speech recognition and spatial audio. To describe room impulse responses, well defined features according to ISO 3382 are typically applied, such as e.g. reverberation time, clarity, strength or the direct-to-reverberant energy ratio. However, are these features sufficient to capture and objectively describe the overall sound quality of a room?In this article, we approached this research question by analyzing the contribution of several acoustic features in a multidimensional feature space for a large data set of RIRs. In addition to classical room acoustic features as defined by ISO 3382, new features adapted from statistics, signal processing and speech analysis, were explored. The room impulse responses under test were represented in this feature space and were clustered based on their distribution in this space. It was found that classical acoustic features play a major role in the clustering structure, while additional descriptors, like statistical moments, provide complementary information, particularly in characterizing the reverberant part of the impulse response. Future work will conduct listening tests to clarify whether RIRs that are grouped together sound alike. Such knowledge is relevant for data-driven acoustics or machine learning applications involving room impulse responses.