Real-Time Evaluation of an RTF-Steered Binaural MVDR Beamformer Incorporating an External Microphone

* Presenting author
Day / Time: 20.03.2019, 11:20-11:40
Room: Saal 5-a
Typ: Poster
Das Posterforum zu diesen Postern findet am Mittwoch von 16:00 - 16:40 Uhr direkt am Poster im ersten Obergeschoss - Posterinsel H - statt. Die Poster hängen von Dienstag morgen bis Donnerstag mittag.
Abstract: Besides suppressing all undesired noise components and preserving the desired source, the goal of binaural noise reduction algorithms is to preserve the so-called binaural cues of both the desired source and the residual noise components. In this contribution, the binaural minimum variance distortionless response (MVDR) beamformer is considered, which is steered using the relative transfer function (RTF) vector of the desired source and thus preserves its binaural cues. In addition to four head-mounted microphones in a binaural hearing aid device we also consider a spatially separated external microphone. In [1] an estimator for the RTF vector of the desired source was proposed that exploits this external microphone, where it was shown that this estimator outperforms state-of-the-art estimators while exhibiting a low computational complexity. In this contribution, a real-time STFT-based implementation of the RTF-steered binaural MVDR beamformer in [1] was evaluated for different reverberant acoustic scenarios, showing that this beamformer yields good results in terms of noise reduction and binaural cue preservation performance.[1] N. Gößling, S. Doclo, RTF-Based Binaural MVDR Beamformer Exploiting an External Microphone in a Diffuse Noise Field, in Proc. ITG Conference on Speech Communication, Oldenburg, Germany, Oct. 2018, pp. 106-110.