Towards a data-driven plane wave decomposition from multichannel room impulse responses (en)

* Presenting author
Day / Time: 08.03.2023, 14:20-14:40
Room: Saal X11-12
Typ: Regulärer Vortrag
Abstract: Plane wave decomposition of microphone array measurements is one prominent solution of the beam forming problem. Finding robust solutions for this typically ill-conditioned inverse problem is challenging, especially when considering full audio bandwidth and very few sensors. In a recent work we approached data-driven learning that utilizes deep models consisting of CNN-LSTM layers initially for a 2D array geometry and horizontal acoustic impulse response data generated by an image source model. For such scenario we have shown that time and direction of arrival of early room reflections can be estimated in a robust sense. In this contribution, we will proceed with 3D array geometry and data. We will discuss dedicated deep model architectures that handle the considerably larger feature space. We pay special attention to the feature engineering as this heavily affects, how model architectures handle the 3D data as quasi images. The influence of practical limitations, like the number of microphones, their self-noise, and spherical grid sampling is investigated to assess the robustness of this 3D approach.


strpos(): Passing null to parameter #1 ($haystack) of type string is deprecated

/kunden/431424_13355/webseiten/subdomains/, line 35
32.              */
33.             if (headers_sent()) {
34.                 $encoding = false;
35.             } elseif (strpos($HTTP_ACCEPT_ENCODING, 'x-gzip') !== false) {
36.                 $encoding = 'x-gzip';
37.             } elseif (strpos($HTTP_ACCEPT_ENCODING, 'gzip') !== false) {
38.                 $encoding = 'gzip';