Neural dynamics of induced throat vibrations during vocal emotion recognition

This study demonstrates that induced throat vibrations bias the categorization of ambiguous emotional vocalizations and modulate early and late ERP components, providing evidence that embodied somatosensory feedback significantly influences vocal emotion processing.

The Gist

Imagine you are trying to guess what emotion someone is feeling just by listening to their voice. Sometimes, the voice is clear (like a loud, angry shout), but other times, it's muddy and confusing (like someone whispering something that sounds halfway between fear and anger).

This study asks a fascinating question: Does the physical feeling of a voice change how we understand it?

Usually, when we listen to someone, we only use our ears. But when we speak, we feel our throat vibrate. The researchers wondered: What if we could trick your brain into feeling those vibrations even when you are just listening? Would your brain use that physical feeling to help decode the emotion?

Here is the story of their experiment, explained simply.

The Setup: The "Vibrating Neck" Experiment

The researchers put a tiny, harmless vibrator on the participants' throats (right where the voice box is). They played them sounds of people saying "Ahhh" in different emotional tones.

• The Sounds: They created "muddy" voices that were a mix of Fear and Anger. Some were mostly angry, some mostly fearful, and some were a perfect 50/50 mix.
• The Trick: While the participants listened, the vibrator on their neck would buzz.
  • Sometimes the buzz matched the voice (e.g., an angry voice with an angry-sounding buzz).
  • Sometimes it clashed (e.g., a fearful voice with an angry buzz).
  • Sometimes there was no buzz at all.

The Results: The Body Helps the Brain

The participants had to quickly decide: "Is this voice Angry or Fearful?"

1. The "Bias" Effect (The Body Talks Back)
The results showed that the vibration acted like a nudge. If the vibrator buzzed with an "Angry" rhythm, people were much more likely to say the voice was angry, even if the voice itself was confusing.

• Analogy: Imagine you are trying to hear a friend in a noisy room. If someone taps your shoulder and points in a specific direction, you are more likely to look that way. The vibration was the "shoulder tap" that told the brain, "Hey, this sounds like anger!"

2. The "Confusion" Effect (When the Body and Brain Disagree)
When the voice was fearful but the vibration was angry, the brain got confused. It had to work harder to figure out what was going on. This showed up in the brain scans as a "glitch" signal, similar to when you hear a sentence that doesn't make sense and your brain goes, "Wait, what?"

3. The Brain's "Speedometer"
The researchers used EEG (brain waves) to see what happened inside the participants' heads. They found two key moments:

• The Early Spark (100-200ms): When the vibration matched the voice, the brain lit up faster. It was like having a turbo boost. The brain recognized the emotion quicker because the physical feeling confirmed what the ears heard.
• The Late Check (400-800ms): When the vibration didn't match the voice, the brain had to do extra work later on to resolve the conflict. It was like a detective realizing the clues don't add up and having to re-investigate the case.

The Big Picture: We Are "Embodied" Listeners

The main takeaway is that we don't just "hear" emotions; we "feel" them.

Think of your brain as a smart home security system.

• The Ears are the cameras watching the door.
• The Throat Vibrations are the motion sensors on the floor.

Usually, the cameras (ears) do the job. But if the footage is blurry (ambiguous voice), the motion sensor (throat vibration) helps the system decide if an intruder is there. If the camera sees a shadow and the floor sensor feels a step, the alarm (emotion recognition) goes off faster and more confidently.

Why This Matters

This study proves that embodied cognition is real. We aren't just passive listeners; our bodies are active participants in understanding what others are feeling.

• For the Future: This could help design better technology for people with hearing impairments (using vibrations to help them "feel" emotions) or improve how we teach robots to understand human feelings.
• The Limitation: The study mostly used female participants and artificial vibrations. Future research needs to see if this works for everyone and with real, natural voice vibrations.

In short: When we listen to emotions, our bodies are whispering secrets to our brains, helping us make sense of the world, especially when the audio is fuzzy.

Technical Summary

1. Problem Statement

While the role of embodied cognition (the simulation of observed actions/sensations) in facial emotion recognition is well-established, its contribution to emotional prosody (vocal emotion) remains underexplored. Specifically, the mechanism by which vibrotactile feedback from the vocal tract influences the perception of vocal emotions is unknown.

• Gap: Speakers experience internal vibrations from their vocal cords during speech, which serve as interoceptive/proprioceptive feedback. However, no study has investigated how induced external throat vibrations in a listener affect the categorization of emotional voices, nor the associated neural dynamics.
• Hypothesis: The authors posit that induced throat vibrations act as a vibrotactile feedback loop. They hypothesized that:
  1. Behaviorally: Vibrations congruent with the voice emotion (e.g., anger vibration + angry voice) would bias perception toward that emotion, particularly for ambiguous stimuli where sensory information is unclear.
  2. Neurally: Congruent and incongruent conditions would modulate specific Event-Related Potential (ERP) components (early sensory processing and late integration) and activate brain regions associated with interoception, somatosensation, and motor simulation (e.g., insula, somatosensory cortex, motor areas).

2. Methodology

Participants:

• $N=24$ healthy adults (18 females, 6 males; mean age 23.67).
• Normal hearing, no neurological/psychiatric history.

Stimuli:

• Auditory: "Aah" vocalizations from the Montreal Affective Voices database.
• Morphing: Voices were morphed between Fear and Anger to create five ambiguity levels (10-90, 40-60, 50-50, 60-40, 90-10).
• Vibrations: Induced using a small vibrator (Ortofon BC-10) taped to the participant's laryngeal prominence (throat).
  • Conditions: Pure Anger vibration, Pure Fear vibration, or No vibration (Control).
  • Timing: Vibrations were delivered simultaneously with the voice stimulus.

Experimental Design:

• Within-subjects design: 5 Emotional Voice levels $\times$ 3 Vibration conditions (Anger, Fear, None).
• Task: Participants categorized the voice as "Anger" or "Fear" as quickly and accurately as possible within a 3-second window.
• Procedure: 6 runs of 113 trials each.

Data Acquisition & Analysis:

• EEG: 64-channel Biosemi ActiveTwo system.
• Preprocessing: Artifact rejection (ICA for blinks/saccades), peak-to-peak amplitude thresholding, averaging into 50ms time bins (0–1000ms post-stimulus).
• Statistical Modeling:
  • Behavior: Generalized Linear Mixed Models (GLMM) for categorical responses; Linear Mixed Models (LMM) for reaction times.
  • EEP: Bayesian General Linear Mixed Models (GLMM) testing interactions between Cluster, Emotional Voice, and Vibration.
  • Source Reconstruction: Performed to localize neural generators of ERP effects.

3. Key Results

A. Behavioral Results

• Main Effect: Induced vibrations significantly skewed emotional ratings.
• Congruency Bias: Participants were more likely to categorize a voice as "Anger" when an Anger vibration was induced, and vice versa for Fear.
• Ambiguity Interaction: The biasing effect was most pronounced for ambiguous voices (40-60% emotion blends).
  • Specifically, for the most ambiguous 50-50 condition, an Anger vibration significantly increased "Anger" responses compared to Fear or No-vibration conditions.
  • This supports the hypothesis that vibrotactile feedback acts as a disambiguating cue when auditory information is uncertain.

B. ERP Results (Frontocentral Cluster C4)

The study identified significant modulations in early and late processing stages:

• Early Processing (100–300ms):
  • N100/P100: Modulated by vibration type, suggesting early attention-driven interoceptive effects.
  • P200: Congruent vibrations facilitated processing (increased amplitude for low-ambiguity anger; decreased for ambiguous fear). This indicates early facilitation of emotion recognition.
• Late Processing (400–900ms):
  • N400/LPC: Significant modulations observed between 500–800ms.
  • Incongruence: Incongruent conditions (e.g., Angry voice + Fear vibration) elicited distinct negative deflections (N400-like) and Late Positive Components (LPC), suggesting error detection, conflict resolution, and integration of conflicting sensory modalities.

C. Source Reconstruction

Source analysis localized the effects to a distributed network involving:

• Interoception & Somatosensation: Insula, Primary Somatosensory Cortex (S1), and Supplementary Motor Area (SMA).
• Motor Simulation: Primary Motor Cortex (M1) and Premotor areas.
• Higher-Order Integration: Dorsolateral Prefrontal Cortex (DLPFC), Orbitofrontal Cortex (OFC), Anterior Cingulate Cortex (ACC), and Superior Parietal Lobule.
• Timing:
  • Early (300-400ms): Anterior Cingulate and Insula activation for incongruent fear.
  • Late (700-950ms): Motor and Parietal areas involved in resolving incongruence and integrating feedback.

4. Key Contributions

1. First Evidence of Vibrotactile Embodiment in Voice: This is the first study to demonstrate that induced throat vibrations in a listener directly influence the perception of vocal emotions, providing empirical support for the "embodied simulation" hypothesis in the auditory domain.
2. Disambiguation Mechanism: The study clarifies that vibrotactile feedback is most critical for ambiguous stimuli, acting as a sensory cue to resolve uncertainty in emotional prosody.
3. Temporal Dynamics: It maps the precise neural timeline of this process, showing that embodiment effects occur as early as 100ms (sensory/attention) and persist through late integration stages (800ms+), involving both facilitation (congruent) and conflict resolution (incongruent) mechanisms.
4. Neural Network Mapping: It identifies a specific network (Insula, S1, M1, DLPFC) that integrates vibrotactile feedback with auditory emotional processing, bridging the gap between interoception and social cognition.

5. Significance and Implications

• Theoretical: The findings strongly support the Simulation of Smiles (SIMS) model extended to voice, suggesting that emotional perception is not purely auditory but relies on the re-experiencing of the physical sensations (vibrations) associated with vocal production.
• Clinical & Applied:
  • Neurological Disorders: Could inform therapies for conditions involving impaired emotional processing (e.g., Autism Spectrum Disorder, Schizophrenia) by utilizing vibrotactile cues to aid emotion recognition.
  • Human-Computer Interaction (HCI): Suggests that haptic feedback (vibration) in VR/AR or telepresence systems could significantly enhance the realism and emotional clarity of synthetic voices.
  • Assistive Technology: Potential applications for hearing-impaired individuals who rely on somatosensory cues to interpret speech and emotion.

Limitations noted by authors: The sample was predominantly female; future studies should test different emotions and use participant-specific induced vibrations for higher ecological validity.