Toward Robust and Adaptive Electroencephalography-based Auditory Attention Decoding for Neuro-steered Hearing Application

다국어 초록 (Multilingual Abstract)

Nowadays, the number of people with hearing loss is increasing rapidly and is estimated to reach around 2.5 billion worldwide by 2050, according to a report from the World Health Organization. Hearing loss severely affects both the physical and mental health of patients. For instance, it has a negative impact on cognitive and language development in infants and children, and it can significantly harm adults’ mental health by causing social isolation and unemployment. To address this issue, one of the current solutions is to provide patients with hearing assistive devices. These devices function by amplifying speech sounds while suppressing background noise. However, in multi-talker situations (commonly referred to as the cocktail party scenario), these devices struggle to detect and amplify the attended speech. To overcome this challenge, Auditory Attention Decoding (AAD) has been introduced, as it can infer the attended speaker by decoding brain signals. Although AAD has been widely studied and developed under laboratory conditions, several limitations still question its feasibility in real-world applications. First, under passive listening conditions, when the listener is not consciously attending to the incoming sound, can the decoding algorithm still operate effectively? Second, most existing AAD algorithms are trained in a supervised manner and remain fixed during operation. This setup may limit their performance in real-world environments, where both the acoustic surroundings and brain signals are highly dynamic. This dissertation addresses these two challenges. First, it investigates the feasibility of using brain signals to decode auditory attention by analyzing neural tracking of the speech envelope under extreme passive listening conditions. Second, it proposes a fast, time-adaptive, and unsupervised AAD algorithm designed for plug-and-play operation. The results of this study demonstrate the feasibility of applying AAD in real-world environments, showing that it can function effectively under everyday listening conditions. Furthermore, the proposed algorithm performs well in a plug-and-play manner, highlighting its superiority and potential for integration into future hearing technologies.

목차 (Table of Contents)

• Abstract VI
• [List of Figures] VIII
• Chapter I General introduction 1
• 1.1. Hearing loss and its neural consequences 2
• 1.2. Hearing assistive devices and Auditory Attention Decoding (AAD) 4
• 1.3. The basis of brain and auditory system 4
• 1.4. Electroencephalography 7
• 1.5. The robustness of neural tracking of speech envelope 9
• 1.6. Current progress in AAD 11
• 1.7. Research goal 12
• Chapter II The robustness of neural tracking of speech 15
• 2.1. Introduction 17
• 2.2. Methods 19
• 2.3. Results 22
• 2.4. Discussion and Conclusion 23
• Chapter III Time adaptive, Unsupervised Auditory Attention Decoding 25
• 3.1. Introduction 27
• 3.2. Methods 29
• 3.3. Results 33
• 3.4. Discussion and Conclusion 41
• Chapter IV Rapid AAD algorithm via a hybrid decoder 43
• 4.1. Introduction 45
• 4.2. Methods 46
• 4.3. Results 50
• 4.4. Discussion and Conclusion 53
• Chapter V General conclusion 56
• 5.1. General conclusions 57
• 5.2. Limitation and future works 58
• References 60