Temporal Coordination of Articulatory and Respiratory Events Prior to Speech Initiation

Oksana Rasskazova, Christine Mooshammer, Susanne FuchsInterspeech 2019

TL;DR

This study investigates the intricate coordination between breathing and speech articulation before we even start speaking. By tracking lung movements and vocal tract actions, researchers found that exhalation onset is tightly linked to the initial consonant's articulation, suggesting a sophisticated, pre-planned motor sequence rather than just a reactive process. This work sheds light on the hidden complexities of speech preparation.

What Problem Does This Paper Address?

We often think of speech as a continuous flow of sound, but the planning and execution of even a single utterance involve a complex interplay of bodily systems. While previous research has explored how articulators (tongue, lips, jaw) prepare for upcoming sounds during silent pauses, the role of respiration – the fundamental engine of speech production – has been largely overlooked. This paper addresses the question of how breathing movements are temporally coordinated with articulatory movements and acoustic events prior to the actual start of speech. Understanding this coordination is crucial for a complete model of speech planning, especially considering that respiration operates on a much slower timescale than articulation.

Methods

The researchers employed a multi-modal approach to capture the fine-grained temporal dynamics of speech preparation. They simultaneously recorded:

  • Electromagnetic Articulography (EMA): This technique tracks the precise movements of articulators like the tongue tip, tongue middle, tongue back, jaw, and lips using small sensors.
  • Inductance Plethysmography: This method measures changes in thoracic (chest) and abdominal volume to monitor breathing patterns (inhalation and exhalation).
  • Acoustics: A standard microphone captured the audio signal.

Data was collected from 12 native German speakers reading sentences. The analysis focused on the time window immediately preceding the acoustic onset of speech. Key events were manually and automatically labelled, including:

  • The beep signal (cue to start reading)
  • Acoustic onset of speech
  • Inhalation and exhalation onsets
  • Lower lip opening onset
  • Tongue tip gesture onset and offset for the initial consonant (/t/ or /n/).

Statistical analysis involved linear mixed-effects models to examine the influence of the initial segment and speaker variability on the timing of these events.

Datasets

The study used a custom dataset of 11 sentences read by 12 native German speakers. The focus was on six target utterances, each repeated five times, containing initial segments of /t/ or /n/. Data from six speakers has been analyzed so far. This is a newly collected dataset for this specific research question.

Key Insights and Ideas

  • Respiration is not passive: The study strongly suggests that breathing is actively coordinated with speech planning, not just a background process.
  • Exhalation is a key anchor: The onset of exhalation appears to be a critical temporal marker, tightly synchronized with the articulatory preparation for the initial consonant.
  • Mouth opening and inhalation are linked: While there's speaker variability, mouth opening during the preparatory phase seems to be related to the inhalation phase, potentially facilitating airflow.
  • Manner of articulation matters: The type of initial consonant (/t/ vs. /n/) influences the precise timing of articulatory and respiratory events, indicating that speech planning is sensitive to phonetic details.
  • Active preparation, not just a byproduct: The observed coordination suggests that articulatory movements before speech are not merely passive consequences of breathing but active, goal-directed preparations for the upcoming utterance.

Key Results

  • Acoustic onset follows exhalation: The acoustic onset of speech occurred, on average, about 69 ms after the onset of exhalation, with no significant difference between /t/ and /n/.
  • Consonant constriction and exhalation are synchronized: For the nasal consonant /n/, the onset of the tongue tip constriction phase was almost synchronous with exhalation onset. For the stop consonant /t/, the constriction onset occurred, on average, 23 ms before exhalation onset.
  • Variability in inhalation and lip opening: The timing between inhalation onset and lower lip opening showed higher variability (up to 44%) compared to the coordination between exhalation and articulatory events.
  • Speaker-specific patterns: Significant speaker-specific differences were observed, particularly in the timing of inhalation relative to mouth opening.

Limitations and Future Work

  • Limited dataset: Only six speakers have been fully analyzed, and the study focused on only two initial consonants (/t/ and /n/).
  • Inhalation onset ambiguity: In some cases, inhalation onset was difficult to precisely determine from the pre-processed signals.
  • Nasal vs. oral breathing: The study could not differentiate between mouth and nasal breathing, which might influence respiratory dynamics.

Future work will involve analyzing the remaining speakers and extending the analysis to other initial segments (/a/, /h/, /f/). The researchers also plan to investigate whether voiceless fricatives exhibit tighter respiratory-articulatory coordination compared to vowels. Further exploration into the role of nasal versus oral breathing is also suggested.

Summary updated

BibTeX
@inproceedings{rasskazova2019temporal,
  author = {Oksana Rasskazova and Christine Mooshammer and Susanne Fuchs},
  title = {{Temporal Coordination of Articulatory and Respiratory Events Prior to Speech Initiation}},
  booktitle = {Interspeech},
  year = {2019},
}
Figure