Sound Source Localization Exploiting Signal and Geometry Conditioning Mechanisms

Beschreibung

Microphone arrays enable direction of arrival (DOA) estimation for applications such as speech communication devices, hearing systems, and acoustic scene analysis. However, reliable DOA estimation remains challenging in realistic environments, where reverberation, background noise, and simultaneously active sources distort the spatial information. Learning-based methods, particularly deep neural network (DNN)-based approaches, can provide robust estimates in such conditions, but commonly require information that identifies the desired source and often depend on a fixed microphone array geometry observed during training. Motivated by these limitations, the first focus is on estimating the DOA of a desired source in scenarios with simultaneously active interfering sources. An auxiliary microphone placed near the desired source is exploited to perform feature conditioning. The second focus is on reducing the dependence of DNN-based estimators on a fixed array geometry. The influence of different integration locations within the network is further investigated.

Autor*in

Ulrik Kowalk

Themen in »Sound Source Localization Exploiting Signal and Geometry Conditioning Mechanisms«

Acoustic Localization Deep Neural Networks Auxiliary Microphone

Stimmen zu »Sound Source Localization Exploiting Signal and Geometry Conditioning Mechanisms«

Details

Link teilen