Dynamic Time-Locking Mechanism in the Cortical Representation of Spoken Words

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Linguistics

Background:

• Human speech conveys complex meanings, but the neural mapping of dynamic acoustic signals to linguistic representations remains unclear.
• Understanding how the brain decodes variable speech signals is essential for cognitive neuroscience and artificial intelligence.

Purpose of the Study:

• To investigate how the brain encodes acoustic-phonetic features of speech using magnetoencephalography (MEG).
• To determine the role of temporal information processing in decoding spoken words from neural activity.
• To compare neural processing of speech versus non-speech sounds with similar temporal characteristics.

Main Methods:

• Utilized physiologically-inspired machine-learning models to map natural acoustic variability to MEG data from 16 healthy volunteers.
• Compared models with different temporal information representations for decoding spoken words.
• Analyzed cortical evoked responses to assess speech encoding fidelity and phoneme content.

Main Results:

• Dynamic time-locking of cortical activation to speech input is critical for encoding acoustic-phonetic features.
• Non-speech environmental sounds, even with similar temporal modulation, did not show similar time-locking.
• The amplitude envelope of speech was well reconstructed from cortical evoked responses, indicating high temporal fidelity.
• Phoneme content was reflected in cortical responses concurrently with spectrotemporal features.

Conclusions:

• Speech is encoded in the brain with exceptional temporal fidelity, involving dynamic time-locking of neural activity.
• This neural tracking mechanism may involve entrainment of cortical oscillations to the speech amplitude envelope.
• The brain instantaneously transforms acoustic features into linguistic representations during speech processing.