Cortical Auditory Attention Decoding During Music and Speech Listening.

Area of Science:

• Neuroscience
• Cognitive Science
• Signal Processing

Background:

• Cortical recordings can identify attended speakers in complex auditory environments.
• Stimulus reconstruction from electroencephalogram (EEG) data approximates attended sound envelopes.
• Previous research primarily focused on auditory attention decoding for speech, with limited exploration in music.

Purpose of the Study:

• To investigate the efficacy of auditory attention detection (AAD) techniques in music listening scenarios.
• To compare AAD performance for music versus speech.
• To explore the role of training data in model construction for AAD.

Main Methods:

• Applied established speech-based AAD techniques to a music listening task with concurrent distraction.
• Utilized linear regression for stimulus reconstruction from EEG data.
• Compared reconstructed sound envelopes with actual stimuli to assess correlation.

Main Results:

• AAD proved successful for both speech and music listening, though reconstruction accuracy differed.
• The importance of specific training data for model performance was emphasized.
• Linear regression models, when trained on musical signals, can decode auditory attention during music listening.

Conclusions:

• Auditory attention detection is feasible for music, extending beyond speech applications.
• Model training data significantly impacts AAD performance.
• Linear regression is a viable method for AAD in music, provided appropriate model training.