Noise-Tagging Auditory Attention Decoding

Recent advancements in neuroscience have demonstrated the feasibility of identifying the target of auditory focus using non-invasive electroencephalography (EEG) recordings. Building on these findings, numerous algorithms fo auditory
attention decoding (AAD) have been suggested. These innovations could pave the way for what are often referred to as neurosteered hearing devices.

In this study, we combined the state-of-the- art AAD with the noise-tagging stimulus protocol. In the visual domain, it has been shown that the code-modulated visual evoked potential (c-VEP) can be decoded from the EEG with extremely high accuracy, leading to high communication rates. We introduced these code-modulated responses to the auditory domain, evoking the code-modulated auditory evoked potentials (c-AEP), by amplitude modulating the speaker-specific audio with pseudo-random noise-codes. By performing this amplitude modulation, each speaker audio is "watermarked" with a unique identifier, that may facilitate the (accuracy and speed of the) decoding of the attended speaker.

We carried out a pilot EEG experiment in which participants were presented with a stream of audio either to the left ear, or to the right ear. Additionally, we performed a pilot experiment in which two streams were simultaneously presented, one to each of the ears. The participant was instructed to attend to one of the two streams. We presented both modulated and unmodulated audio.