A Binaural Model that Analyses Acoustic Spaces and Stereophonic Reproduction Systems by Utilizing Head Rotations

Exploiting Deep Neural Networks and Head Movements for Robust Binaural Localization of Multiple Sources in Reverberant Environments

“…The proposed approach to exploiting head movements is based on late information fusion -the information from the model predictions is integrated. This is in contrast to the approach in [12] which adopted early fusion at the feature level by averaging cross-correlation patterns across different head orientations. Late fusion is preferred here for a couple of reasons: i) the use of head rotation is not needed during model training and thus it is more straightforward to generate data for training robust localisation models (DNNs); ii) early feature fusion tends to lose information which can otherwise be exploited by the system.…”

“…Late fusion is preferred here for a couple of reasons: i) the use of head rotation is not needed during model training and thus it is more straightforward to generate data for training robust localisation models (DNNs); ii) early feature fusion tends to lose information which can otherwise be exploited by the system. As a result, the proposed system is able to deal with overlapping sound sources in reverberant conditions, while the system reported in [12] was tested in anechoic conditions with a single source.…”

Sound source localization is a multisystem process

“…1 is not the only way to explain the integration of auditoryspatial and head-position cues [1] nor is it a formal model. Such models would require connecting auditory-spatial cues to head-centric angle, head-position cues to head angle, and specifying a method of integrating the cues [3]. If the auditory spatial cues are interaural time (ITD) or level (ILD) differences, then two head-centric angles produce the same ITD and (probably) ILD ( Fig.…”

Audio-motor integration for robot audition

“…Both ILD and IPD are known to be subject-dependent and frequency-dependent cues. This is captured by the so-called head related transfer functions (HRTFs) determined by the shape Audio-motor integration [8,9,20,30,31,33,34,38,41,48,55,60,61] Active audition [4,6,39,40,45,67]…”

“…Note that the above mentioned studies rely on static localization techniques applied from different viewpoints. In contrast, [8] computes the average binaural cross-correlation of a continuously rotating head to estimate a source's azimuth in an anechoic scenario. To the best of the authors' knowledge, exploiting continuous sensor movement for 2D sound sound direction estimation with a single microphone, as showed in Section 2.3 of this chapter, has not been proposed before.…”