gordo2023unsupervised.pdf (695.72 kB)

Unsupervised Accuracy Estimation for Brain-Computer Interfaces based on Auditory Attention Decoding

Download (695.72 kB)
posted on 2023-08-15, 13:48 authored by Miguel A. Lopez-Gordo, Simon GeirnaertSimon Geirnaert, Alexander Bertrand

Auditory attention decoding (AAD) algorithms process brain data such as electroencephalography (EEG) in order to decode to which of multiple competing sound sources a person attends. Example use cases are neuro-steered hearing aids or brain-computer interfaces (BCI) for patients with severe motor or cognitive impairments. Recently, it has been shown that it is possible to train such AAD decoders in an unsupervised setting, where there is no ground truth available regarding which of the sound sources is attended. However, in the absence of such ground-truth labels, it is difficult to quantify the accuracy of the decoders, in particular when working with patients who can neither communicate nor show any evidence of collaboration. In this paper, we use principles of digital communications to estimate the AAD accuracy during a competing talker listening task in a completely unsupervised manner. We show that our proposed unsupervised estimation technique can accurately determine the accuracy of the decoder in a transparent-for-the-user way, for different amounts of training data, decision window lengths, and amounts of estimation data. Furthermore, since different applications demand different targeted accuracies, our approach can estimate the required minimal amount of training required for any given target accuracy. In neuro-steered hearing aids, our approach could support time-adaptive decoding, dynamic gain control, and neurofeedback training. In BCIs, it could support a robust communication paradigm for caregivers and professionals to reach and gain insights into people who cannot communicate by any means.


grant B-TIC-352-UGR20 FEDER/Junta de Andalucía-Council for Economic Transformation, Industry, Knowledge and Universities

grant PID2021-128529OA-I00, MCIN/AEI/10.13039/501100011033, by ERDF A way of making Europe

grant PROYEXCEL_00084 funded by Consejería de Universidad, Investigación e Innovación, Junta de Andalucía (2021)

PDMT1/22/009, KU Leuven

FWO project nr. G081722N

the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 802895)

the Flemish Government (AI Research program)


Email Address of Submitting Author

ORCID of Submitting Author


Submitting Author's Institution

KU Leuven

Submitting Author's Country

  • Belgium