Neural oscillations track natural but not artificial fast speech: Novel insights from speech-brain coupling using MEG

Ana-Sofía Hincapié Casas, Tarek Lajnef, Annalisa Pascarella, Hélène Guiraud-Vinatea, Hannu Laaksonen, Dimitri J. Bayle, Karim Jerbi, Véronique Boulenger

TL;DR

This study reveals that our brains synchronize neural oscillations to the rhythm of natural speech, but not to artificially sped-up speech, even when the syllable rate is the same. This suggests that the brain's tracking of speech is sensitive to the naturalness of the speech, not just its temporal structure, highlighting the importance of natural speech in understanding speech perception.

What Problem Does This Paper Address?

Understanding how our brains process spoken language is a complex challenge, partly due to the vast variability in speech. A key aspect of this is how the brain tracks the rhythm of speech, particularly the syllable rate. While previous research has shown that neural oscillations can synchronize with speech, most studies have used artificially sped-up speech (time-compressed). This leaves a significant gap: how does the brain track natural variations in speech rate, and how does this differ from tracking artificially altered speech? This paper aims to bridge this gap by directly comparing how our brains synchronize to natural fast speech versus time-compressed speech.

Methods

The researchers used Magnetoencephalography (MEG) to record brain activity from 24 participants. They presented participants with different types of auditory stimuli:

  • Natural Normal Speech: Sentences spoken at a typical pace.
  • Natural Fast Speech: Sentences spoken at a faster, but still natural, pace.
  • Time-Compressed Speech: Sentences from the normal pace that were artificially sped up to match the rate of the natural fast speech.
  • Amplitude-Modulated Noise: Control stimuli that mimicked the temporal envelope of the speech but lacked linguistic content.

The core analysis involved cortico-acoustic coherence, which measures how well the brain's electrical activity (neural oscillations) synchronizes with the amplitude envelope of the speech signal. They specifically looked at frequencies corresponding to the syllable rates of the speech stimuli. To ensure the findings were robust, they compared the coherence during speech presentation to control conditions: shuffled data (where the speech and brain activity were mismatched) and pre-stimulus baseline data. They also performed seed-based phase-coupling analyses to examine network dynamics.

Datasets

The study used a custom-created dataset of 288 meaningful French sentences. These sentences were recorded by a professional actor at both normal and fast rates. Time-compressed versions of these sentences were generated, and amplitude-modulated noise stimuli were created using the envelopes of the original sentences. No pre-existing benchmark datasets were used for training or evaluation in the traditional machine learning sense.

Key Insights and Ideas

  • Naturalness Matters: The brain's ability to synchronize neural oscillations to speech rhythm is not solely dependent on the syllable rate. The naturalness of the speech significantly influences this synchronization.
  • Motor Cortex Involvement: The motor cortex, typically associated with speech production, plays a crucial role in tracking natural fast speech, showing stronger coupling than to artificially sped-up speech. This suggests a sensorimotor integration process.
  • Beyond Canonical Frequencies: The study emphasizes the importance of looking at stimulus-specific frequencies (tied to syllable rate) rather than relying on broad frequency bands (like theta) that might miss crucial dynamics.
  • Articulatory Features: The stronger coupling to natural fast speech might reflect the brain's ability to track specific articulatory features that are characteristic of natural speech production, which are altered in time-compressed speech.

Key Results

  • Natural Speech Tracking: Cortico-acoustic coherence was significantly increased in auditory and (pre)motor cortices for both natural normal speech (at ~6.25 Hz) and natural fast speech (at ~8.75 Hz), with the frequency shifting to match the syllable rate.
  • Artificial Speech Failure: Time-compressed speech, despite having the same syllable rate as natural fast speech (~8.75 Hz), did not show significant cortico-acoustic coupling at that frequency.
  • Motor Cortex Specificity: Direct contrasts revealed significantly stronger coupling in the right articulatory/motor cortex for natural fast speech compared to both time-compressed and natural normal speech at ~8.75 Hz.
  • Network Connectivity: Enhanced phase synchronization was observed between the right motor cortex and left temporo-parietal and motor regions for natural fast speech compared to time-compressed speech.
  • Linguistic Content: Amplitude-modulated noise stimuli, even with speech-like envelopes, did not elicit significant cortico-acoustic coupling, suggesting that linguistic content plays a role.

Limitations and Future Work

  • Sample Size: While 24 participants is a reasonable number for MEG studies, the authors note that some uncorrected statistics in supplementary analyses (e.g., for fronto-temporal networks in time-compressed speech) did not survive correction for multiple comparisons across a large number of sources.
  • Stimulus Specificity: The study used single sentences. Future work could investigate brain coupling to longer extracts of natural speech at various rates.
  • Methodological Constraints: The authors acknowledge that using natural speech is more methodologically constraining than using artificial stimuli.
  • Future Directions: The authors suggest future research should continue to use stimulus-specific frequencies and natural speech material. They also propose investigating the developmental trajectory of neural tracking of speech in children. Additionally, comparing brain oscillatory responses to unintelligible natural speech with spectro-temporally complex intelligible speech is suggested to disentangle the roles of linguistic content versus acoustic complexity.

Summary updated

Abstract

Neural oscillations contribute to speech parsing via cortical tracking of hierarchical linguistic structures, including syllable rate. While the properties of neural entrainment have been largely probed with speech stimuli at either normal or artificially accelerated rates, the important case of natural fast speech has been largely overlooked. Using magnetoencephalography, we found that listening to naturally-produced speech was associated with cortico-acoustic coupling, both at normal (∼6 syllables/s) and fast (∼9 syllables/s) rates, with a corresponding shift in peak entrainment frequency. Interestingly, time-compressed sentences did not yield such coupling, despite being generated at the same rate as the natural fast sentences. Additionally, neural activity in right motor cortex exhibited stronger tuning to natural fast rather than to artificially accelerated speech, and showed evidence for stronger phase-coupling with left temporo-parietal and motor areas. These findings are highly relevant for our understanding of the role played by auditory and motor cortex oscillations in the perception of naturally produced speech.
BibTeX
@article{casas2021neural,
  author = {Ana-Sofía Hincapié Casas and Tarek Lajnef and Annalisa Pascarella and Hélène Guiraud-Vinatea and Hannu Laaksonen and Dimitri J. Bayle and Karim Jerbi and Véronique Boulenger},
  title = {{Neural oscillations track natural but not artificial fast speech: Novel insights from speech-brain coupling using MEG}},
  journal = {NeuroImage},
  year = {2021},
  doi = {10.1016/j.neuroimage.2021.118577},
}
Figure