Abstractness in speech-metronome synchronisation: P-centres as cyclic attractors

Plínio A. Barbosa, Pablo Arantes, Alexsandro R. Meireles, Jussara M. VieiraInterspeech 2005

TL;DR

This paper proposes a novel dynamical systems perspective on speech rhythm, viewing "p-centers" (perceptual anchors of syllables) not as fixed points but as emergent properties of an underlying "phrase stress oscillator." By analyzing Brazilian Portuguese speakers synchronizing with a metronome, they demonstrate that p-center timing is influenced by both the acoustic properties of consonants and the metronome's rate, suggesting a flexible, predictive mechanism for speech timing.

What Problem Does This Paper Address?

The paper tackles the long-standing puzzle of "p-centers" in speech perception and production. While it's known that listeners and speakers often anchor their timing to something other than the precise start of a syllable (the p-center), the underlying mechanism and the factors influencing its location have remained debated. Previous research often focused on static acoustic features or treated p-centers as fixed points. This paper argues that this view is too simplistic and proposes that p-centers are dynamic, emergent phenomena arising from the brain's predictive modeling of speech rhythm, akin to how a pendulum settles into a stable swing. They aim to bridge the gap between the abstract nature of temporal prediction in speech and the concrete acoustic realization of syllables.

Methods

The authors employ a speech-metronome synchronization task with a Brazilian Portuguese speaker. Participants were asked to produce sequences of identical CV syllables in time with a metronome at various rates. Crucially, they captured both the metronome signal and the participant's speech simultaneously.

The core of their analysis involves:

  • P-center Calculation: P-centers were computed by projecting the metronome pulse onto the speech signal, treating the metronome pulse as the predicted anchor point.
  • Acoustic Analysis: They analyzed acoustic features of the syllables, specifically focusing on:
  • The rise time of total energy at the CV transition (hypothesizing that steeper rises would lead to p-centers closer to the vowel onset).
  • The energy in the high-frequency band of the consonant (hypothesizing that higher high-frequency energy would lead to earlier p-centers). This was quantified as the difference in dB between the 1500 Hz-Nyquist frequency band and the 0-1500 Hz band.
  • Dynamical Systems Concepts: They frame the synchronization task using concepts from dynamical systems theory, particularly "frequency synchronization" (matching rates) and "phase synchronization" (stable timing relationship between p-centers and vowel onsets). They also introduce the idea of a "bifurcation" in behavior when the metronome rate crosses a certain threshold.

What's novel here is the explicit application of a dynamical systems framework to p-center phenomena, moving beyond static correlations to explore emergent timing behaviors and the concept of an underlying "phrase stress oscillator."

Datasets

The paper does not describe a formal dataset in the traditional machine learning sense. Instead, it presents data from a single Brazilian Portuguese speaker performing a controlled experimental task. The "dataset" consists of 21 different sequences of identical CV syllables produced at four different metronome rates. The syllables were chosen to vary in consonant acoustic properties (energy rise, high-frequency content) and vowel aperture.

Key Insights and Ideas

  • P-centers as Cyclic Attractors: The paper's central idea is that p-centers are not fixed points but rather the observable manifestation of an underlying, dynamically oscillating system that predicts upcoming syllable onsets.
  • Predictive Timing Mechanism: The findings suggest that speakers actively predict upcoming vowel onsets and use this prediction to synchronize with external cues like a metronome. This prediction is influenced by acoustic cues.
  • Consonant Acoustics Matter: The spectral tilt (high-frequency energy) of consonants significantly influences p-center timing, with more high-frequency energy leading to earlier p-centers relative to the vowel onset. This aligns with the idea that certain acoustic features might "pull" the perceived timing.
  • Rate-Dependent Behavior (Bifurcation): The synchronization behavior changes qualitatively as the metronome rate increases, suggesting a shift in the underlying cognitive process. This "bifurcation" indicates that the system doesn't scale linearly with tempo.
  • Coupled Oscillator Model: The results support the authors' previous proposal of a coupled-oscillator model for speech rhythm, where phrase stress oscillators interact with external timing signals.

Key Results

  • Spontaneous Rate Synchronization: At the spontaneous metronome rate (108 bpm), acoustic features like consonant spectral tilt and energy rise at the vowel onset showed significant correlations with p-center phase. Specifically, consonant spectral tilt had a -57% correlation with phase.
  • Effect of Metronome Rate:
    • At slower rates (80 bpm and 108 bpm), phase synchronization was generally good, with p-centers often aligning predictably with vowel onsets based on acoustic cues.
  • At faster rates (138 bpm and 208 bpm), phase synchronization broke down significantly. The metronome pulse "slid" along the syllable envelope, indicating a loss of stable timing.
  • Consonant Influence: Syllables starting with fricatives like [s] and [ʃ] (high-frequency consonants) showed earlier p-centers, supporting hypothesis 2. Consonants like [p] (lower energy) and vowels like [a] (higher energy rise) also showed predictable patterns.
  • Vowel Duration Adjustment: The subject significantly shortened vowel duration as the metronome rate increased, a key adaptation strategy.

Limitations and Future Work

  • Single Speaker: The study relies on data from a single speaker, limiting the generalizability of the findings. Future work should involve multiple speakers from different linguistic backgrounds.
  • Limited Syllable Set: While 21 CV syllables were used, the complexity of natural speech is far greater. Expanding the syllable inventory to include codas and more complex structures would be beneficial.
  • Focus on Production: The study primarily examines speech production. Further research could explore how these findings relate to speech perception and the interaction between production and perception.
  • Abstractness of Time: The paper highlights the "abstractness of time relations" needed for synchronization. Future work could delve deeper into the cognitive mechanisms underlying this abstract representation and prediction.
  • Cross-Linguistic Comparison: The authors suggest their findings support a coupled-oscillator model that can explain cross-linguistic variability. Further studies comparing different languages would be valuable.

Summary updated

BibTeX
@inproceedings{barbosa2005abstractness,
  author = {Plínio A. Barbosa and Pablo Arantes and Alexsandro R. Meireles and Jussara M. Vieira},
  title = {{Abstractness in speech-metronome synchronisation: P-centres as cyclic attractors}},
  booktitle = {Interspeech},
  year = {2005},
}
Figure