Auxiliary function-based algorithm for blind extraction of a moving speaker

Abstract

In this paper, we propose a novel algorithm for blind source extraction (BSE) of a moving acoustic source recorded by multiple microphones. The algorithm is based on independent vector extraction (IVE) where the contrast function is optimized using the auxiliary function-based technique and where the recently proposed constant separating vector (CSV) mixing model is assumed. CSV allows for movements of the extracted source within the analyzed batch of recordings. We provide a practical explanation of how the CSV model works when extracting a moving acoustic source. Then, the proposed algorithm is experimentally verified on the task of blind extraction of a moving speaker. The algorithm is compared with state-of-the-art blind methods and with an adaptive BSE algorithm which processes data in a sequential manner. The results confirm that the proposed algorithm can extract the moving speaker better than the BSE methods based on the conventional mixing model and that it achieves improved extraction accuracy than the adaptive method.

Double Nonstationarity: Blind Extraction of Independent Nonstationary Vector/Component from Nonstationary Mixtures -- Algorithms

“…Surprising results are obtained by CSV-AuxIVE since it shows improvement with growing α similarly to the methods employing nonstationarity. This is in contrast with the fact that the method comes from the optimization of non-Gaussianitybased source model [25], [48]. The theoretical explanation of this behavior goes beyond the scope this paper.…”

“…The nonstationary mixing process is less frequently described by a more specific parameterization such as that used in [46], [47]. Recently, semi-time-variant models denoted as CMV and CSV (Constant Mixing/Separating Vector) have been considered in [14], [16], [48]. CMV and CSV are designed for BSE/jBSE in which the SOI is static or moving, respectively, on a dynamic background.…”

“…Note that the separating vectors w k are independent of t while the mixing vectors a k,t depend on it. This parameterization corresponds to the semi-time-variant CSV mixing model advocated in [14], [16], [48].…”

“…We have compared eight algorithms. BOGIVE w [48], [56], FastDIVA from [14], and CSV-AuxIVE [48] represent non-Gaussianity-based methods that inherently work with the hypothesis that L = 1. In BOGIVE w and FastDIVA, the rational nonlinearity φ(s) = s * 1+|s| 2 denoted as "rati" is used; CSV-AuxIVE utilizes the standard nonlinearity for super-Gaussian sources [25].…”

Mask-based Neural Beamforming for Moving Speakers with Self-Attention-based Tracking

“…Source movements are also a problem for microphonearray-based blind source separation. Recently, several studies proposed estimating time-invariant separation filters that are robust to source movements [33]- [35]. However, these approaches may deal with only relatively small source movements because the filters are time-invariant.…”