Acoustic and Articulatory Study of Ewe Vowels: A Comparative Study of Male and Female

Kowovi Comivi Alowonou, Jianguo Wei, Wenhuan Lu, Zhicheng Liu, Kiyoshi Honda, Jianwu DangInterspeech 2019 · OpenReview

Abstract

In order to investigate the difference in Ewe males and Ewe females during the production of Ewe vowels, results from the comparative quantitative and qualitative assessments of tongue shape and movement using ultrasound imaging as well as the comparative evaluation of F1 and F2 frequency values from data collected from 9 Ewe male speakers and 6 Ewe female speakers, were presented in this study. The results showed that vowels are produced with higher formant frequencies by Ewe female speakers compared to Ewe male speakers, except for the vowel /ε/ produced with a lower F1 frequency by Ewe females. The articulatory results showed a higher and more forwarder tongue configuration for Ewe male compared to female counterparts.

Introduction

Most speech production research has historically focused on a small subset of the world's languages, leaving the physiological mechanisms of languages from developing countries largely unmapped. Ewe, a language spoken primarily in Togo, Ghana, and Benin, is one such language. Understanding the relationship between articulatory movements (how the tongue moves) and acoustic outputs (the resulting sound) is crucial for both theoretical linguistics and practical applications like Automatic Speech Recognition (ASR). This paper provides the first quantitative comparative study of male and female Ewe speakers using modern imaging techniques.

Methodology

The researchers employed a multimodal approach to capture speech data from 15 native Ewe speakers (9 male, 6 female) from Togo. The stimuli consisted of 21 tokens of CV sequences using the seven Ewe oral vowels: /a, e, ε, i, o, ɔ, u/.

Acoustic Analysis

Audio recordings were analyzed using Linear Predictive Coding (LPC) via PRAAT software to estimate the first two formants (F1 and F2), which correspond to vowel height and backness. To account for anatomical differences, frequencies were normalized using the Lobanov z-score transformation.

Articulatory Imaging

A Telemed Echo Blaster ultrasound system with a convex probe was used to visualize the tongue's surface in the sagittal plane at 95 frames per second. Statistical modeling included Smoothing Spline ANOVA (SSANOVA) to compare tongue shape curves and Linear Mixed Effects (LME) models to quantify gender-based differences in displacement.

Results and Discussion

Acoustic Vowel Space

The study found significant gender-based differences in acoustic output. For most vowels, females exhibited significantly higher F1 and F2 frequencies. For instance, the F2 difference for /o/ was estimated at a massive 986.09 Hz. As shown in the acoustic vowel space plot, the female vowel space is not only shifted but expanded relative to the male space, moving "downward" (higher F1) and "forward" (higher F2).

Acoustic vowel space comparison between Ewe male and female speakers.

One notable exception was the vowel /ε/, which females produced with a lower F1 frequency than males. This deviation from the general trend suggests unique constriction patterns in the lower pharynx or specific lip dynamics for this vowel in Ewe.

Articulatory Tongue Configuration

In the articulatory domain, Ewe males generally showed a higher and more forward tongue configuration compared to their female counterparts. The LME models highlighted a significant horizontal difference for the vowel /i/, where the male tongue was positioned approximately 5.99 mm further forward than the female tongue.

SSANOVA plots comparing tongue shapes for Ewe oral vowels across genders.

These results generally support phonetic theories linking tongue raising to decreased F1 and tongue advancement to increased F2. However, the less clear mapping for vowels like /i/ and /ε/ suggests that other articulators, such as the jaw or lips, may play a more significant role in these instances.

Conclusion

This research highlights the physiological and acoustic distinctions between male and female Ewe speakers. While females produce vowels with higher formant frequencies and a larger acoustic space—likely due to smaller laryngeal and pharyngeal tracts—males utilize a higher and more forward tongue position.

Future work should expand the sample size and include Ewe's nasal vowels to provide a more comprehensive mapping of the language's phonology. Integrating these findings into speech processing services could significantly improve ASR performance for Ewe speakers.

Summary updated

Abstract

In order to investigate the difference in Ewe males and Ewe females during the production of Ewe vowels, results from the comparative quantitative and qualitative assessments of tongue shape and movement using ultrasound imaging as well as the comparative evaluation of F1 and F2 frequency values from data collected from 9 Ewe male speakers and 6 Ewe female speakers, were presented in this study. The results showed that vowels are produced with higher formant frequencies by Ewe female speakers compared to Ewe male speakers, except for the vowel /ε/ produced with a lower F1 frequency by Ewe females. The articulatory results showed a higher and more forwarder tongue configuration for Ewe male compared to female counterparts.
BibTeX
@inproceedings{alowonou2019acoustic,
  author = {Kowovi Comivi Alowonou and Jianguo Wei and Wenhuan Lu and Zhicheng Liu and Kiyoshi Honda and Jianwu Dang},
  title = {{Acoustic and Articulatory Study of Ewe Vowels: A Comparative Study of Male and Female}},
  booktitle = {Interspeech},
  year = {2019},
  url = {https://openreview.net/forum?id=vEUAb9fl-Bs},
}
Figure