Speaker Identity is Robustly Encoded in Spatial Patterns of Intracranial EEG for Attention Decoding.

Area of Science:

• Neuroscience
• Auditory Perception
• Brain-Computer Interfaces

Background:

• The human auditory cortex tracks attended speech, but how speaker identity is encoded spatially remains unclear.
• Existing auditory attention decoding (AAD) methods primarily rely on temporal speech dynamics.

Purpose of the Study:

• To investigate if speaker identity is encoded in time-invariant spatial patterns within the auditory cortex.
• To develop a novel AAD framework based on spatial speaker identification.

Main Methods:

• Utilized human intracranial EEG (iEEG) to record neural activity.
• Developed a novel AAD framework focusing on spatial activation maps ('neural fingerprints') for speaker identification.
• Decoupled the decoding of speaker identity ('who') from speech timing ('when').

Main Results:

• Demonstrated that speaker identity is reliably reflected in distinct, time-invariant spatial activation maps in the auditory cortex.
• Achieved state-of-the-art speech extraction using the spatial identification AAD framework, outperforming temporal models in short time windows.
• Observed dynamic shifts in spatial neural codes corresponding to attentional switches.

Conclusions:

• Speaker identity is a robust, spatially distributed feature in the auditory cortex.
• Spatial speaker identification offers a high-speed, complementary approach to temporal decoding for AAD.
• Findings support the development of advanced neuro-steered hearing technologies.