This paper considers the auditory attention detection (AAD) paradigm, where the goal is to determine which of two simultaneous speakers a person is attending to. The paradigm relies on recordings of the listener's brain activity, e.g., from electroencephalography (EEG). To perform AAD, decoded EEG signals are typically correlated with the temporal envelopes of the speech signals of the separate speakers. In this paper, we study how the inclusion of various degrees of auditory modelling in this speech envelope extraction process affects the AAD performance, where the best performance is found for an auditory-inspired linear filter bank followed by power law compression. These two modelling stages are computationally cheap, which is important for implementation in wearable devices, such as future neuro-steered auditory prostheses. We also introduce a more natural way to combine recordings (over trials and subjects) to train the decoder, which reduces the dependence of the algorithm on regularization parameters. Finally, we investigate the simultaneous design of the EEG decoder and the audio subband envelope recombination weights vector using either a norm-constrained least squares or a canonical correlation analysis, but conclude that this increases computational complexity without improving AAD performance.

This work shows the importance of using realistic binaural listening conditions and training on a balanced set of experimental conditions to obtain results that are more representative for the true AAD performance in practical applications.

When multiple people talk simultaneously, the healthy human auditory system is able to attend to one particular speaker of interest. Recently, it has been demonstrated that it is possible to infer to which speaker someone is attending by relating the neural activity, recorded by electroencephalography (EEG), with the speech signals. This is relevant for an effective noise suppression in hearing devices, in order to detect the target speaker in a multi-speaker scenario. Most auditory attention detection algorithms use a linear EEG decoder to reconstruct the attended stimulus envelope, which is then compared to the original stimuli envelopes to determine the attended speaker. Classifying attention within a short time interval remains the main challenge. We present two different convolutional neural network (CNN)-based approaches to solve this problem. One aims to select the attended speaker from a given set of individual speaker envelopes, and the other extracts the locus of auditory attention (left or right), without knowledge of the speech envelopes. Our results show that it is possible to decode attention within 1-2 seconds, with a median accuracy around 80%, without access to the speech envelopes. This is promising for neuro-steered noise suppression in hearing aids, which requires fast and accurate attention detection. Furthermore, the possibility of detecting the locus of auditory attention without access to the speech envelopes is promising for the scenarios in which per-speaker envelopes are unavailable. It will also enable establishing a fast and objective attention measure in future studies. Index TermsConvolutional neural networks (CNN), auditory attention detection (AAD), electroencephalography (EEG), neurosteered auditory prosthesis, brain-computer interface (BCI)

This work shows how the background noise level and relative positions of competing talkers impact attention decoding accuracy. It indicates in which circumstances a neuro-steered noise suppression system may need to operate, in function of acoustic conditions. It also indicates the boundary conditions for the operation of EEG-based attention detection systems in neuro-steered hearing prostheses.