Modeling source-tract interaction in speech production: voicing onset vs. vowel height after a voiceless obstruent

Jorge C. Lucero, Laura L. Koenig, Susanne FuchsInterspeech 2012

TL;DR

This paper investigates why higher vowels require more intraoral pressure to initiate voicing after a voiceless consonant. They propose and model that the acoustic coupling between the vocal folds and the vocal tract, specifically the interaction of vocal fold vibration frequency with the first formant, explains this phenomenon. This work offers a theoretical framework for understanding a subtle but consistent aspect of speech production.

What Problem Does This Paper Address?

When we speak, the vibration of our vocal folds (voicing) and the shape of our vocal tract (which determines vowel sounds) are intimately linked. Researchers observed a puzzling pattern: it takes more pressure in the mouth (intraoral pressure, Pio) to start vocal fold vibration when producing high vowels (like /i/ or /i/) compared to low vowels (like /a/), even though high vowels are typically produced with a higher fundamental frequency (fo), which usually requires less pressure to initiate. This contradicts expectations based on basic phonation principles. The paper aims to explain this discrepancy by exploring how the vocal tract's acoustics might influence the pressure needed to start vocal fold vibration.

Methods

The authors employ a theoretical model that combines two key components:

  1. Vocal Fold Oscillation Model: They use Titze's mucosal wave model, a well-established biomechanical model of vocal fold tissue, to describe how vocal folds vibrate. This model considers factors like mass, damping, stiffness, and the pressure driving the vibration.
  2. Vocal Tract Model: The vocal tract is simplified as a hard-walled tube of a specific length. The crucial aspect here is how sound waves reflect from the open end of the tract back towards the vocal folds, influencing the glottal airflow and pressure.

The core of their approach is to model the interaction between these two components. They derive an equation for the transglottal pressure (the pressure difference across the vocal folds) required to initiate voicing. This equation explicitly includes terms related to the vocal fold vibration frequency ($f_0$) and the first formant ($F_1$) of the vocal tract. They then fit this theoretical model to experimental data to see if it can reproduce the observed pressure differences.

Datasets

The study uses data collected from eight adult speakers of Standard German. These speakers produced consonant-vowel sequences (with voiceless obstruents followed by /a/, /i/, or /1/) embedded in carrier sentences. The intraoral pressure, vocal fold frequency ($f_0$), and first formant ($F_1$) were measured. A total of 762 tokens were analyzed after data cleaning. This dataset appears to be a re-analysis of data previously collected for another study [1].

Key Insights and Ideas

  • Acoustic Coupling is Key: The paper strongly suggests that the acoustic coupling between the vocal folds and the vocal tract is the primary reason for the higher voicing onset pressure observed for high vowels.
  • "U"-Shaped Relationship: The model predicts a "U"-shaped relationship between the transglottal pressure needed for voicing onset and the ratio of vocal fold frequency to the first formant ($f_0/F_1$). This means there's an optimal ratio where voicing onset pressure is minimized.
  • Vowel Height and $f_0/F_1$: High vowels, with their lower $F_1$, tend to have a higher $f_0/F_1$ ratio compared to low vowels. When this ratio falls on a specific side of the "U" (the side where increasing the ratio increases pressure), it explains the observed data.
  • Inertive Load: When $f_0$ is below $F_1$, the vocal tract acts as an "inertive load," which can actually decrease the pressure needed for phonation. However, the paper's model shows that the reflected wave from the vocal tract can counteract this, leading to increased pressure requirements at certain $f_0/F_1$ ratios.

Key Results

  • The theoretical model, when fitted to the data, reproduces the observed trend of higher intraoral pressure for high vowels compared to low vowels.
  • The model successfully predicts the difference in mean transglottal pressure between low and high vowels (DP) for most subjects, showing good agreement with the experimental data.
  • For most subjects, the model's predicted $f_0/F_1$ ratio for minimum transglottal pressure aligns with the observed data patterns, suggesting the "U"-shaped relationship is valid.
  • Minor discrepancies were noted for subjects W2 and M4, where the model produced flatter patterns than the data suggested, indicating potential limitations in the simplified model or the data interpretation for these specific cases.

Limitations and Future Work

  • Simplified Vocal Tract Model: The vocal tract is modeled as a simple tube, and vowel variations are characterized solely by the first formant ($F_1$). This ignores the complex geometry and higher formants of the vocal tract.
  • Fixed Glottal Configuration: The model assumes a fixed glottal configuration for each subject, which is contradicted by the observed "intrinsic $f_0$ effect" (where $f_0$ varies with vowel height).
  • Future Work: The authors suggest lifting these simplifications in future studies. This includes incorporating more detailed vocal tract models and accounting for dynamic changes in glottal configuration to better understand the extent to which acoustic coupling alone explains the observed phenomenon. They also note that their work aligns with recent theories on nonlinear source-filter coupling in phonation.

Summary updated

BibTeX
@inproceedings{lucero2012modeling,
  author = {Jorge C. Lucero and Laura L. Koenig and Susanne Fuchs},
  title = {{Modeling source-tract interaction in speech production: voicing onset vs. vowel height after a voiceless obstruent}},
  booktitle = {Interspeech},
  year = {2012},
}
Figure