Acoustic correlates of glottal gaps

Gang Chen, Jody Kreiman, Yen-Liang Shue, Abeer AlwanInterspeech 2011

TL;DR

This paper investigates the acoustic consequences of incomplete vocal fold closure (glottal gaps or GGs) during speech. By combining high-speed laryngeal imaging with acoustic analysis, the researchers found that the size of these gaps significantly impacts voice quality, particularly by increasing spectral noise and influencing spectral tilt, though these effects can be complex and depend on other vocal parameters like pitch.

What Problem Does This Paper Address?

When we speak, our vocal folds vibrate to produce sound. Ideally, they close completely during each vibration cycle. However, sometimes they don't fully close, leaving a small opening called a glottal gap (GG). While GGs are known to occur in voice disorders, they also happen in normal speech, especially in women. Previous research hinted that GGs might be linked to acoustic features like "breathiness" and spectral tilt (how quickly high frequencies decrease in the voice signal), but these connections weren't systematically studied, especially concerning the size of the gap. This paper aims to fill that gap by quantitatively measuring GG size and exploring its relationship with various acoustic measures.

Methods

The core of this research lies in the innovative combination of two data acquisition techniques:

  • High-Speed Laryngeal Imaging: This allows researchers to capture the vocal folds vibrating at thousands of frames per second. From these videos, they can precisely measure the glottal area waveform for each vocal fold cycle.
  • Synchronous Audio Recording: Simultaneously, the sound produced is recorded.

From the glottal area waveforms, the researchers derived a novel measure called Relative Glottal Gap Area (RGGA). This metric quantifies the GG size relative to the maximum glottal area within a cycle, providing a normalized measure of how incomplete the closure is. They also calculated standard measures like Open Quotient (OQ) and Asymmetry Quotient (AQ).

The acoustic analysis involved extracting several key voice source parameters:

  • Cepstral Peak Prominence (CPP) and Harmonics-to-Noise Ratio (HNR): These measures relate to the periodic structure and spectral noise level of the voice, respectively.
  • Spectral Tilt Measures (H*₁ - H*₂, H*₁ - A*₃): These quantify how quickly the energy in the voice signal decreases with increasing frequency, often related to breathiness and spectral characteristics.

Statistical analyses (correlations and regression) were used to determine the relationships between RGGA and these acoustic measures.

Datasets

The study utilized data from six subjects (three female, three male) who provided sustained vowel sounds (/i/) under various conditions of pitch (low, normal, high) and voice quality (breathy, normal, pressed). The data was previously collected for another study [8]. For Experiment 2, a single female speaker performed a glide phonation, transitioning from breathy to pressed voice.

Key Insights and Ideas

  • RGGA as a Key Metric: The introduction of RGGA provides a quantitative way to assess glottal gap size, moving beyond qualitative descriptions or simple presence/absence.
  • GG Size and Noise: Larger glottal gaps (higher RGGA) are consistently associated with increased spectral noise (lower CPP and HNR). This aligns with the perception of breathiness.
  • Complex Relationship with Spectral Tilt: The effect of GG size on spectral tilt (H*₁ - A*₃) is not straightforward. While one experiment suggested a negative correlation (larger gap, smaller tilt), another, under steady pitch conditions, showed a positive correlation (larger gap, larger tilt). This highlights the interplay of factors.
  • Pitch Matters: Higher pitches tend to be associated with larger glottal gaps, particularly in female speakers. This might be due to increased vocal fold tension at higher pitches.
  • GGs as an "Additional Degree of Freedom": The presence of GGs, especially in female speakers, introduces more variability in voice production and acoustic correlates, acting as an extra factor influencing voice quality beyond standard parameters like OQ and AQ.

Key Results

  • Voice Quality and Pitch Effects: Breathy phonations and high-pitched phonations showed significantly larger RGGA compared to normal/pressed and normal/low-pitched phonations, respectively.
  • Correlation with Acoustic Measures (Experiment 1):
    • RGGA was positively correlated with F0 (pitch).
  • RGGA was negatively correlated with CPP, HNR, and H*₁ - A*₃.
  • Correlation with Acoustic Measures (Experiment 2 - Glide Phonation):
    • RGGA was strongly negatively correlated with AQ and positively with H*₁ - H*₂.
  • RGGA was strongly positively correlated with H*₁ - A*₃.
  • Predictive Power: In the context of GGs, RGGA and AQ were found to be better predictors of H*₁ - H*₂ than OQ, suggesting GGs introduce new acoustic dynamics.

Limitations and Future Work

  • Limited Sample Size: The study involved only six subjects, and Experiment 2 used only one speaker. This limits the generalizability of the findings.
  • Complexity of Interactions: The paper acknowledges that the acoustic consequences of GGs are "quite complicated" and depend on various factors like pitch, voice quality, and individual speaker characteristics.
  • Future Work: The authors suggest that more data is needed to generalize these results. Further research could explore the perceptual impact of these acoustic correlates and investigate a wider range of phonatory modes and speaker populations. The study also hints at the need for more sophisticated models that can account for the complex interplay of factors influencing voice production when GGs are present.

Summary updated

BibTeX
@inproceedings{chen2011acoustic,
  author = {Gang Chen and Jody Kreiman and Yen-Liang Shue and Abeer Alwan},
  title = {{Acoustic correlates of glottal gaps}},
  booktitle = {Interspeech},
  year = {2011},
}
Figure