Acoustic cues of vowel quality to coda nasal perception in southern Min

Ying Chen, Vsevolod Kapatsinski, Susan Guion-AndersonInterspeech 2012

TL;DR

This study investigates how vowel sounds influence the perception of nasal consonants at the end of syllables in Southern Min. Researchers found that vowel quality significantly impacts listeners' ability to distinguish between different nasal sounds, and that specific acoustic features within the vowel provide crucial cues for this perception. This work sheds light on the complex interplay between vowels and consonants in speech perception and has implications for understanding sound change.

What Problem Does This Paper Address?

Speech sounds don't exist in isolation; they influence and are influenced by their neighbors. This phenomenon, known as coarticulation, means that the acoustic properties of a sound can be altered by the sounds that come before or after it. This paper focuses on how vowels, specifically their quality (e.g., high front vs. low back), affect the perception of nasal consonants (like 'm', 'n', 'ng') when they appear at the end of a syllable (coda position). While it's known that vowels can carry cues for consonant perception, the precise acoustic mechanisms and their impact on specific languages like Southern Min, which retains certain historical nasal contrasts lost in other Chinese dialects, are not fully understood. The authors aim to explore: 1) the inherent confusability of different coda nasals, 2) how vowel quality impacts this perception, and 3) which specific acoustic features within the vowel are most informative for identifying the nasal consonant.

Methods

The researchers conducted a perceptual confusion experiment using synthesized speech stimuli. They selected eight monosyllabic words from Quanzhou Southern Min, each ending in one of the three nasal consonants (/m/, /n/, /ŋ/) and preceded by one of three vowel types (/i/, /ə/, /a/). These words were recorded by four native speakers.

For the perceptual experiment, 160 target tokens (8 words x 5 repetitions x 4 talkers) were presented to 15 native Southern Min listeners. The listeners were tasked with identifying the final nasal consonant as either 'm', 'n', or 'ng' in a forced-choice task. The audio stimuli were RMS normalized and embedded in pink noise to control for intensity variations and simulate a more realistic listening environment.

An acoustic analysis of the vowel formants (F1, F2, F3) at their midpoint and endpoint, as well as the change in these formants during the vowel, was performed for the recorded stimuli. This analysis aimed to identify acoustic correlates of the perceptual findings.

Datasets

The study utilized a custom dataset of eight monosyllabic words from Quanzhou Southern Min. These words were chosen to represent combinations of three coda nasal types (/m/, /n/, /ŋ/) and three vowel types (/i/, /ə/, /a/), with a mid-level tone. The dataset was recorded by four native speakers, resulting in 160 unique tokens for the perceptual experiment. No pre-existing datasets were used for training or evaluation.

Key Insights and Ideas

  • Vowel Quality is a Major Player: The study strongly confirms that the vowel preceding a coda nasal significantly influences how accurately listeners can identify that nasal. This isn't just a minor effect; it can lead to substantial misperceptions.
  • High Front Vowels are Tricky: The high front vowel /i/ was found to be particularly challenging, leading to more misidentifications of coda nasals compared to mid (/ə/) and low (/a/) vowels. This suggests that the acoustic environment created by /i/ obscures the cues for nasal place of articulation.
  • Acoustic Cues Within the Vowel Matter: The research highlights that specific acoustic features within the vowel itself, rather than just the overall vowel sound, provide crucial information for nasal perception. Higher formant frequencies at the vowel-nasal boundary and greater formant changes from the vowel's midpoint to its end were associated with more accurate nasal identification.
  • Perceptual Hierarchy Holds: The study found a perceptual hierarchy for coda nasals in Southern Min: /m/ was the most confusable, followed by /ŋ/, and then /n/ was the most accurately identified. This aligns with findings in other languages, suggesting some universal principles in nasal perception.
  • Phonotactics Less Influential Than Acoustics: Even when a specific coda nasal-vowel combination was phonotactically unlikely in Southern Min (e.g., /ən/), listeners still often identified the nasal based on acoustic cues, suggesting that acoustic salience can sometimes override phonotactic constraints.

Key Results

  • The perceptual confusion experiment revealed that /m/ was the most confusable coda nasal, followed by /n/, and then /n/.
  • The high front vowel /i/ led to more misidentifications of coda nasals than the mid vowel /ə/ and low vowel /a/.
  • A significant interaction was found between nasal type and vowel type in perception, indicating that the effect of vowel quality varies depending on the nasal consonant.
  • Acoustic analysis showed that higher formant frequencies at the vowel-nasal juncture and greater formant change from the vowel midpoint to endpoint correlated positively with accurate nasal identification. Specifically, F1 midpoint, F1 endpoint, F1 change, F2 change, F3 endpoint, and F3 change were all significantly correlated with identification accuracy.

Limitations and Future Work

  • Limited Stimuli: The study used a relatively small set of eight words and only three vowel types. Future work could expand this to include more vowels, consonants, and syllable structures.
  • Absence of /cən/: The exclusion of the /cən/ word due to its absence in the Southern Min sound system meant that a complete ANOVA across all three vowels and three nasals was not possible.
  • Focus on Vowel Acoustics: The current study primarily focused on vowel formant structures as acoustic cues. The authors acknowledge that other acoustic features, such as vowel formant amplitude and nasal peaks, could also play a role and are areas for future investigation.
  • Connection to Sound Change: While the study touches upon the historical retention of nasals in Southern Min, it did not directly investigate the link between these perceptual findings and language change. Future research could explore how these perceptual biases might drive or reflect historical sound shifts.

Summary updated

BibTeX
@inproceedings{chen2012acoustic,
  author = {Ying Chen and Vsevolod Kapatsinski and Susan Guion-Anderson},
  title = {{Acoustic cues of vowel quality to coda nasal perception in southern Min}},
  booktitle = {Interspeech},
  year = {2012},
}
Figure