TL;DR
This paper investigates how vocal fold vibration patterns (voice quality and glottal gaps) influence acoustic measures used to analyze speech. By combining high-speed laryngeal imaging with acoustic analysis, they found that measures like open quotient and spectral tilt are significantly affected by both voice quality and incomplete glottal closure, offering a more direct link between physiology and acoustics.
What Problem Does This Paper Address?
Understanding the human voice source is crucial for speech analysis, particularly for diagnosing vocal cord issues and interpreting voice quality. While acoustic measures like spectral tilt (e.g., H1-H2) and cepstral peak prominence (CPP) are commonly used as proxies for voice source characteristics, their direct relationship to the underlying physiological movements of the vocal folds has been unclear. Previous studies often relied on indirect evidence or theoretical models. This paper aims to bridge this gap by directly measuring vocal fold dynamics using high-speed imaging and correlating these physiological parameters with established acoustic measures.
Methods
The researchers used high-speed laryngeal imaging to capture vocal fold vibrations during sustained vowels. From these images, they extracted glottal area waveforms, which were then averaged to represent a single phonation cycle. These averaged waveforms were fitted to a four-parameter voice source model, yielding parameters like the open quotient (OQ), asymmetry coefficient (α), and the speeds of vocal fold opening and closing (Sop and Scp). Concurrently, acoustic measures such as H1-H2, spectral tilt measures (H1-A1, H1-A2, H1-A3), CPP, and harmonic-to-noise ratios (HNR) were calculated from the audio recordings. Statistical analyses, specifically two-way ANOVA, were employed to determine the significance of voice quality (pressed, normal, breathy) and the presence/absence of glottal gaps on these parameters and measures. Correlation analyses were also performed to explore relationships between the model parameters and acoustic measures.
Datasets
The study utilized data from six subjects (3 male, 3 female) who produced sustained vowels (/i/) while varying fundamental frequency (Fo) and voice quality (pressed, normal, breathy). Synchronous audio and high-speed laryngeal imaging were recorded. The most stable 1-second segments of phonation were extracted for analysis. No new benchmark datasets were introduced; rather, existing data was re-analyzed with a focus on direct physiological-acoustic correlations.
Key Insights and Ideas
- Direct Physiological Link: The study provides empirical evidence directly linking vocal fold dynamics (measured via glottal area waveforms) to acoustic measures, moving beyond purely theoretical or indirect correlations.
- Voice Quality's Dominance: Voice quality, particularly breathiness, significantly impacts both voice source parameters (like OQ and α) and acoustic measures (like CPP and HNR), suggesting breathiness introduces more spectral noise and alters vocal fold closure patterns.
- Glottal Gaps Matter: The presence of glottal gaps (incomplete vocal fold closure) demonstrably affects the open quotient (OQ) and the speed of vocal fold opening (Sop), indicating that even subtle physiological deviations can be acoustically detectable.
- H1-H2's Dual Nature: The acoustic measure H1-H2 is shown to be correlated with both the open quotient (OQ) and the asymmetry coefficient (α), aligning with theoretical predictions but also highlighting the complex interplay of these physiological factors.
- Sop's Surprising Sensitivity: While Sop wasn't significantly affected by voice quality, it was significantly influenced by glottal gaps, suggesting that the process of opening might be more sensitive to closure irregularities than the overall duration of opening.
Key Results
- OQ: Significantly affected by both voice quality (pressed < normal < breathy) and glottal gaps (lower with gaps).
- α (Asymmetry): Primarily affected by voice quality, with breathy phonations showing the lowest asymmetry.
- Sop (Opening Speed): Significantly affected by glottal gaps (slower opening with gaps).
- Acoustic Measures: CPP and HNR measures were significantly lower for breathy phonations, indicating more spectral noise. H1-H2 was significantly larger for breathy phonations. Spectral tilt measures (H1-A2, H1-A3) were smallest for pressed and largest for breathy phonations.
- Correlations: H1-H2 showed a strong positive correlation with OQ (r=0.656) and a negative correlation with α (r=-0.473). CPP was negatively correlated with OQ and positively with α.
Limitations and Future Work
- Normalization Effects: The paper acknowledges that normalization of glottal area waveforms might have removed information about glottal area peaks and DC offsets, which represent maximum opening and incomplete closure levels, respectively. Preliminary analysis on a glide phonation suggested these normalized parameters might not be significantly affected by changes in voice quality, but this requires further investigation.
- Scp's Puzzling Behavior: The parameter Scp (closing speed) showed no meaningful correlations with OQ, despite Sop being correlated with OQ. This lack of correlation requires further exploration.
- Limited Subject Pool: The study used a limited number of subjects, and results were not separated by gender due to this limitation. Future work could expand the subject pool to investigate gender-specific differences.
- Further Exploration of Correlations: The paper suggests that the correlations between OQ and Sop, and the lack of correlation between OQ and Scp, are puzzling and warrant further research to fully understand the underlying mechanisms.
Summary updated