This paper considers the auditory attention detection (AAD) paradigm, where goal is to determine which of two simultaneous speakers a person attending to. The paradigm relies on recordings listener's brain activity, e.g., from electroencephalography (EEG). To perform AAD, decoded EEG signals are typically correlated with temporal envelopes speech separate speakers. In this paper, we study how inclusion various degrees modelling in envelope extraction process affects AAD performance, best...

Objective: The electroencephalogram (EEG) is an essential neuro-monitoring tool for both clinical and research purposes, but susceptible to a wide variety of undesired artifacts.Removal these artifacts often done using blind source separation techniques, relying on purely data-driven transformation, which may sometimes fail sufficiently isolate in only one or few components.Furthermore, some algorithms perform well specific artifacts, not others.In this paper, we aim develop generic EEG...

Wireless microphone networks or so-called wireless acoustic sensor (WASNs) are a next-generation technology for audio acquisition and processing. As opposed to traditional arrays that sample sound field only locally, often at large distances from the relevant sources, WASNs allow use many more microphones cover area of interest. However, design such is very challenging, especially real-time signal enhancement due significant data traffic in network. There need scalable solutions, both on...

We introduce a distributed adaptive algorithm for linear minimum mean squared error (MMSE) estimation of node-specific signals in fully connected broadcasting sensor network where the nodes collect multichannel signal observations. assume that to be estimated share common latent subspace with dimension is small compared number available channels at each node. In this case, can significantly reduce required communication bandwidth and still provide same optimal MMSE estimators as centralized...

Objective: We aim to extract and denoise the attended speaker in a noisy two-speaker acoustic scenario, relying on microphone array recordings from binaural hearing aid, which are complemented with electroencephalography (EEG) infer of interest. Methods: In this study, we propose modular processing flow that first extracts two speech envelopes recordings, then selects envelope based EEG, finally uses inform multichannel separation denoising algorithm. Results: Strong suppression interfering...

Objective.A listener's neural responses can be decoded to identify the speaker person is attending in a cocktail party environment.Such auditory attention detection methods have potential provide noise suppression algorithms hearing devices with information about attention.A challenge effect of and other acoustic conditions that reduce accuracy.Specifically, impact ability segregate sound sources perform selective attention, as well external signal processing necessary decode effectively.The...

In a multi-speaker scenario, the human auditory system is able to attend one particular speaker of interest and ignore others. It has been demonstrated that it possible use electroencephalography (EEG) signals infer which someone attending by relating neural activity speech signals. However, classifying attention within short time interval remains main challenge. We present convolutional network-based approach extract locus (left/right) without knowledge envelopes. Our results show decode...

We study the problem of distributed least-squares estimation over ad hoc adaptive networks, where nodes have a common objective to estimate and track parameter vector. consider case there is stationary additive colored noise on both regressors output response, which results in biased local estimators. Assuming that covariance can be estimated (or known priori), we first propose bias-compensated recursive algorithm (BC-RLS). However, this bias compensation increases variance or mean-square...

In this paper, we revisit an earlier introduced distributed adaptive node-specific signal estimation (DANSE) algorithm that operates in fully connected sensor networks. the original algorithm, nodes update their parameters a sequential round-robin fashion, which may yield slow convergence of estimators, especially so when number network is large. When all simultaneously, adapts more swiftly, but can no longer be guaranteed. Simulations show then often gets locked suboptimal limit cycle. We...

Objective.We consider the problem of auditory attention detection (AAD), where goal is to detect which speaker a person attending to, in multi-speaker environment, based on neural activity.This work aims analyze influence head-related filtering and ear-specific decoding performance an AAD algorithm.Approach.We recorded high-density EEG 16 normal-hearing subjects as they listened two speech streams while tasked attend either their left or right ear.The attended ear was switched between...

Total least squares (TLS) is a popular solution technique for overdetermined systems of linear equations, where both the right-hand side and input data matrix are assumed to be noisy. We consider TLS problem in an ad hoc wireless sensor network, each node collects observations that yield node-specific subset equations. The goal compute full set equations distributed fashion, without gathering all these fusion center. To facilitate use dual-based subgradient algorithm (DBSA), we transform...

We present a distributed adaptive node-specific signal estimation (DANSE) algorithm that operates in wireless sensor network with tree topology. The extends the DANSE for fully connected networks, as described previous work. It is argued why topology natural choice if not connected. If desired signals share common latent subspace, it shown converges to same linear MMSE solutions obtained centralized version of algorithm. computational load then shared between different nodes network, and...

Concealable, miniaturized electroencephalography (mini-EEG) recording devices are crucial enablers toward long-term ambulatory EEG monitoring. However, the resulting miniaturization limits inter-electrode distance and scalp area that can be covered by a single device. The concept of wireless sensor networks (WESNs) attempts to overcome this limitation placing multitude these mini-EEG at various locations. We investigate whether optimizing WESN topology compensate for effects in an auditory...

Objective: Noise reduction algorithms in current hearing devices lack informationabout the sound source a user attends to when multiple sources are present. To resolve this issue, they can be complemented with auditory attention decoding (AAD) algorithms, which decode using electroencephalography (EEG) sensors. State-of-the-art AAD employ stimulus reconstruction approach, envelope of attended is reconstructed from EEG and correlated envelopes individual sources. This however, performs poorly...

Change point detection (CPD) aims to locate abrupt property changes in time series data. Recent CPD methods demonstrated the potential of using deep learning techniques, but often lack ability identify more subtle autocorrelation statistics signal and suffer from a high false alarm rate. To address these issues, we employ an autoencoder-based methodology with novel loss function, through which used autoencoders learn partially time-invariant representation that is tailored for CPD. The...

Measurement of neural tracking natural running speech from the electroencephalogram (EEG) is an increasingly popular method in auditory neuroscience and has applications audiology. The involves decoding envelope signal EEG signal, calculating correlation with audio stream that was presented to subject. Typically systems 64 or more electrodes are used. However, practical applications, set-ups fewer required. Here, we determine optimal number electrodes, best position place a limited on scalp....

Objective. To develop an efficient, embedded electroencephalogram (EEG) channel selection approach for deep neural networks, allowing us to match the target model, while avoiding large computational burdens of wrapper approaches in conjunction with networks. Approach. We employ a concrete selector layer jointly optimize EEG and network parameters. This uses Gumbel-softmax trick build continuous relaxations discrete parameters involved process, them be learned end-to-end manner traditional...

Abstract Objective. Spatial auditory attention decoding (Sp-AAD) refers to the task of identifying direction speaker which a person is attending in multi-talker setting, based on listener’s neural recordings, e.g. electroencephalography (EEG). The goal this study thoroughly investigate potential biases when training such Sp-AAD decoders EEG data, particularly eye-gaze and latent trial-dependent confounds, may result models that decode or trial-specific fingerprints rather than spatial...

Abstract Object recognition and categorization are essential cognitive processes which engage considerable neural resources in the human ventral visual stream. However, tuning properties of stream neurons for object shape category virtually unknown. We performed large-scale recordings spiking activity Lateral Occipital Complex response to stimuli dimension was dissociated from dimension. Consistent with studies nonhuman primates, neuronal representations were primarily shape-based, although...

The benefit of using external acoustic sensor nodes for noise reduction in hearing aids is demonstrated a simulated scenario with multiple sound sources. A distributed adaptive node-specific signal estimation (DANSE) algorithm, that has reduced communication bandwidth and computational load, evaluated. Batch-mode simulations compare the performance centralized multi-channel Wiener filter (MWF) DANSE. In scenario, DANSE observed not to be able achieve same as its MWF equivalent, although...

In this paper, we consider the linearly constrained distributed adaptive node-specific signal estimation (LC-DANSE) algorithm, which generates a minimum variance (LCMV) beamformer, i.e., with linear constraints, at each node of wireless sensor network. The algorithm significantly reduces number signals that are exchanged between nodes, and yet obtains optimal LCMV beamformers as if has access to all in We case where steering vectors known, well blind beamforming not known. formally prove...

This article explained how nodes in a network graph can infer information about the topology or its related properties, based on in-network distributed learning, i.e., without relying an external observer who has complete overview over network. Some key concepts from field of SGT were reviewed, with focus those that allow for simple implementation, eigenvector Katz centrality, algebraic connectivity, and Fiedler vector. paper also themselves quantify their individual network-wide influence,...