Combining predictive coding and neural oscillations enables online syllable recognition in natural speech

Area of Science:

• Neuroscience
• Computational Linguistics
• Auditory Processing

Background:

• Natural speech comprehension involves segmenting acoustic signals into linguistic units.
• Theta-gamma oscillation coupling is hypothesized to parse syllables and encode neural activity.
• Speech perception relies on contextual cues for predicting structure and content.

Purpose of the Study:

• To investigate the role of theta-gamma coupling in bottom-up and top-down processing during online syllable identification.
• To develop a computational model simulating syllable recognition in continuous speech.

Main Methods:

• Designed the Precoss (predictive coding and oscillations for speech) computational model.
• The model integrates spectro-temporal syllable representations and theta oscillations for prediction and signaling.
• Evaluated syllable recognition based on theta-gamma coupling effectiveness.

Main Results:

• Syllable recognition accuracy is maximized when theta-gamma coupling aligns spectro-temporal predictions with acoustic input.
• The model demonstrates effective recognition of syllable sequences in continuous speech.
• Theta-gamma coupling is crucial for temporal alignment in speech processing.

Conclusions:

• Neurocomputational modeling successfully integrates predictive coding and neural oscillations to explain online sensory processing in speech.
• Theta-gamma coupling plays a key role in dynamic speech comprehension.
• The findings offer insights into the neural mechanisms underlying real-time speech perception.