Acoustic and Electroglottographic Study of Breathy and Modal Vowels as Produced by Heritage and Native Gujarati Speakers

Kiranpreet NaraInterspeech 2017

TL;DR

This study investigates how heritage and native Gujarati speakers produce breathy and modal vowels, revealing that native speakers exhibit clearer distinctions using a wider range of acoustic and articulatory features. Heritage speakers show a less robust contrast, suggesting potential language attrition effects on phonation.

What Problem Does This Paper Address?

Many languages worldwide utilize different voice qualities beyond the standard "modal" voice, including breathy or creaky sounds. Gujarati, an Indo-Aryan language, is one such example, making phonemic distinctions between breathy and modal vowels. While previous research has explored the acoustic and articulatory differences between these phonation types in Gujarati, this study addresses a crucial gap: understanding how heritage speakers, who learned Gujarati as their first language but are now primarily English speakers, produce these sounds compared to native speakers. This comparison is vital for understanding language acquisition, attrition, and the preservation of phonological features in heritage language speakers.

Methods

The researchers employed a combination of acoustic and electroglottographic (EGG) measurements to analyze vowel production.

  • Acoustic Analysis: Standard acoustic parameters were measured, including:
  • Harmonic-to-Noise Ratio (HNR): A measure of the noisiness of the voice.
  • Cepstral Peak Prominence (CPP): Another measure related to voice periodicity and noisiness.
  • Spectral Tilt Measures (H1-H2, H2-H4, H1-A1, H1-A2, H1-A3): These parameters quantify the relative amplitudes of different harmonics and formants, reflecting how quickly the sound's intensity drops off with increasing frequency.
  • Fundamental Frequency (F0): The basic pitch of the voice.
  • Electroglottographic (EGG) Analysis: EGG measures the electrical impedance between the vocal folds, providing insights into their contact patterns. Key EGG parameters analyzed were:
  • Contact Quotient (CQ): The proportion of the glottal cycle during which the vocal folds are in contact.
  • Derivative-EGG Closure Peak Amplitude (DECPA): A measure of the speed of vocal fold closure.

To analyze the temporal dynamics of these parameters, each vowel was divided into nine sections. Smoothing Spline ANOVA (SS-ANOVA) was used to examine global differences across the entire vowel duration and to determine if differences were statistically significant between the two phonation types and speaker groups.

Datasets

The study utilized a custom dataset of 40 target words in Gujarati, specifically designed to elicit breathy and modal vowel contrasts. These words contained pairs of modal and breathy vowels (/a/ vs /a/, /e/ vs /e/, /o/ vs /o/) in specific contexts. The target vowel was the first vowel following an unaspirated consonant, and subsequent vowels and consonants were also unaspirated to isolate the breathy/modal contrast.

Key Insights and Ideas

  • Native speakers show a robust contrast: Native Gujarati speakers clearly distinguish breathy and modal vowels using a significant number of acoustic and EGG parameters.
  • Heritage speakers' contrast is weaker: While heritage speakers do show a distinction between breathy and modal vowels, this contrast is less pronounced and less consistent across parameters and time compared to native speakers.
  • Parameter differences: Specific acoustic parameters like H2-H4, H1-A1, and H1-A2, along with HNR and CPP, were key differentiators for native speakers. For heritage speakers, the significant differences were primarily observed for H1-A2, CPP, and CQ.
  • Temporal localization: The differences between breathy and modal vowels were not uniform across the entire vowel duration but were often localized to specific parts of the vowel, particularly the middle or latter two-thirds.
  • H1-H2 not a significant differentiator in this study: Interestingly, the H1-H2 measure, often cited as a key indicator of breathy voice in other languages, did not reach statistical significance in distinguishing phonation types for either speaker group in this study, though it did trend in the expected direction.

Key Results

  • Native Speakers: Significant differences between breathy and modal vowels were found for H2-H4 (intervals 3-9), H1-A1 (2-7), H1-A2 (2-8), HNR (4-8), and CPP (3-8). Contact Quotient (CQ) was also significantly different (1-7).
  • Heritage Speakers: Significant differences were observed for H1-A2 (4-7), CPP (4-7), and CQ (3-6). The magnitude of these differences was generally smaller than for native speakers.
  • F0 and DECPA: Fundamental frequency (F0) did not distinguish phonation types for either group. DECPA also did not differentiate phonation types for either speaker group.

Limitations and Future Work

  • Limited Sample Size: The study included 13 native and 6 heritage speakers. A larger sample size, particularly for heritage speakers, could provide more robust statistical power.
  • Heritage Speaker Proficiency: The paper notes that heritage speakers' productions were "like those of the study of Western Armenian where heritage speakers produced more English-like vowel qualities." This suggests potential language attrition. Future work could investigate how different levels of heritage speaker proficiency correlate with phonation production.
  • Perception Study: The authors suggest that future research should also explore the perception of phonation by both heritage and native speakers to understand how these acoustic differences are interpreted.
  • Broader Language Context: While this study focuses on Gujarati, future work could explore similar contrasts in other languages with breathy phonation to gain a broader understanding of cross-linguistic phonetics.

Summary updated

BibTeX
@inproceedings{nara2017acoustic,
  author = {Kiranpreet Nara},
  title = {{Acoustic and Electroglottographic Study of Breathy and Modal Vowels as Produced by Heritage and Native Gujarati Speakers}},
  booktitle = {Interspeech},
  year = {2017},
}
Figure