Robust Full-Sphere Binaural Sound Source Localization

Abstract

We propose a novel method for full-sphere binaural sound source localization that is designed to be robust to real world recording conditions. A mask is proposed that is designed to remove diffuse noise and early room reflections. The method makes use of the interaural phase difference (IPD) for lateral angle localization and spectral cues for polar angle localization. The method is tested using different HRTF datasets to generate the test data and training data. The method is also tested with the presence of additive noise and reverberation. The method outperforms the state of the art binaural localization methods for most testing conditions.

Robust Full-sphere Binaural Sound Source Localization Using Interaural and Spectral Cues

“…In reverberant environments, the total impulse response is comprised of the HRIR and an additional component provided by the acoustic reflections, ζ (t). The acoustic reflections consist of early reflections, which have directionality and later reflections which are diffuse [2,7]. If this sound is recorded at the ears, the recording equipment and procedure may also introduce convolutive noise, ν ζ (t) and additive noise, χ ζ (t).…”

“…To account for phase circularity, the PDF is estimated using all aliases of the IPD in the range (−3π, 3π] for each time-frequency unit. In [2], univariate kernel density estimation was used to estimate the PDF of the IPD in each frequency band. For the proposed method, it was found that the IPD of the direct component of the sound was better estimated by using bivariate kernel density estimation to estimate the PDF of the IPD as a function of frequency:…”

“…This component has such a slow variation that it does not obscure the location of the peaks and notches or the relative level of neighbouring peaks and notches. In order to compare the relative positions of the peaks and notches in the test sound to those in the HRTF templates then, this slow varying component needs to be removed [2,14]. The real cepstrum of the signal, ψ(n) is given by: cv = F −1 {Ψ(p)}, where v ∈ {0, ..., N − 1} are the indices of the cepstral coefficients, cv in the cepstral domain.…”

“…Therefore, in order to localize a non-moving sound source on the full-sphere, the listener's unique HRTF dataset is needed. The method in this paper is based on the method in [2]. The proposed method uses the interaural phase difference (IPD) to estimate the lateral angle of the sound source.…”

Binaural sound localization based on deep neural network and affinity propagation clustering in mismatched HRTF condition

“…Pang et al [1] put forward with reverberation weighting and a more generalized parametric model to further improve the localization performance in the reverberant and noisy environments. A full-sphere binaural localization method is proposed in [11], which applies the Interaural Phase Difference (IPD) for lateral localization and spectral cues for polar angle localization. Although the HRTFs for the training and test sets are captured in different rooms, the models of the dummy head are the same.…”