Aerodynamics of fricative production in european portuguese

Cátia M. R. Pinho, Luis M. T. Jesus, Anna BarneyInterspeech 2009

TL;DR

This paper investigates the aerodynamic and acoustic characteristics of voiced fricative production in European Portuguese, focusing on how voicing is maintained or lost. The authors introduce a novel method for identifying "weak voicing" based solely on oral airflow, which is crucial for understanding speech impairments and developing better rehabilitation strategies.

What Problem Does This Paper Address?

Understanding how voiced fricatives are produced and why they sometimes become devoiced is a significant challenge in speech science. This is particularly relevant for individuals with laryngeal impairments, such as unilateral vocal fold paralysis (UVFP), who struggle to maintain voicing. Current methods for classifying voicing rely on a combination of acoustic and electroglottographic (EGG) signals, but the relationship between EGG signals and the physical activity in the larynx during weak voicing is not well understood. This paper aims to characterize the aerodynamic and acoustic features of voiced fricatives in European Portuguese, with a long-term goal of improving therapy for voice rehabilitation by establishing a baseline of normal speech production.

Methods

The researchers employed a multimodal approach, combining aerodynamic and electroglottographic (EGG) recordings. They analyzed the oral airflow and EGG signals from four native European Portuguese speakers (two male, two female) producing specific speech corpora. The corpus included isolated words with voiced fricatives (/v/, /z/, /Z/) in various positions and these words embedded in carrier sentences.

Key methodological contributions include:

  • Aerodynamic and EGG Recording: Standard techniques were used, including a Rothenberg mask for oral airflow measurement and an EGG processor.
  • Novel "Weak Voicing" Definition: The authors developed a new criterion for identifying "weak voicing" that relies solely on the oral airflow signal. This addresses the limitations of previous methods that struggled with the subtle oscillations observed during weak voicing and the unclear relationship with EGG signals. The criterion involves analyzing the amplitude of oral flow oscillations and comparing them to vowel levels, setting a threshold for "weak voicing."
  • Data Analysis: Speech segments (phones) preceding, during, and following the fricative were analyzed using 20ms windows. Measures included absolute mean oral flow, oral flow waveform amplitude, fundamental frequency (F0), and F1 amplitude. Relative changes in these parameters between phones were calculated.
  • Annotation: Recordings were manually annotated using Praat to identify phone boundaries.

Datasets

The study utilized a custom-collected dataset of European Portuguese speech. This dataset consisted of:

  • 9 isolated words containing the voiced fricatives /v/, /z/, and /Z/ in initial, medial, and final positions.
  • 42 different carrier sentences embedding these same 9 words.

This corpus was designed to capture a variety of consonantal and vocalic contexts surrounding the target fricatives.

Key Insights and Ideas

  • Voicing is Dynamic in EP Fricatives: European Portuguese voiced fricatives exhibit a high rate of devoicing or weak voicing, making them ideal for studying the onset and maintenance of voicing.
  • Oral Airflow as a Primary Indicator: The study successfully demonstrated that oral airflow alone can be a reliable indicator of weak voicing, circumventing the ambiguities associated with EGG signals in these subtle phonetic contexts.
  • Aerodynamic Reductions During Fricatives: All analyzed aerodynamic parameters (oral airflow, F0, and F1 amplitude) showed reductions during fricative production compared to adjacent vowel sounds, regardless of speaker gender.
  • Gender Differences in Airflow: Female speakers generally exhibited lower absolute oral airflow values than male speakers during fricative production.
  • "Weak Voicing" is Prevalent: The newly defined "weak voicing" was found to be a common phenomenon in the corpus, with significant percentages of fricatives classified as such, particularly for /z/ and /Z/.

Key Results

  • Significant Reduction in Oral Flow: A decrease in oral flow amplitude during fricatives compared to adjacent vowels was observed across all speakers.
  • F0 and F1 Amplitude Decreases: Mean F0 values decreased during fricative production for both male and female speakers. Similarly, F1 amplitude showed a decrease between vowel and fricative segments.
  • Prevalence of Weak Voicing: Based on the 60% oral flow amplitude threshold, the study found high percentages of weak voicing: /v/ ≥ 88%, /z/ ≥ 71%, and /Z/ ≥ 59% for male speakers, and /v/ ≥ 71%, /z/ ≥ 82%, and /Z/ ≥ 53% for female speakers. These figures are comparable to or higher than previous devoicing rates reported for EP.
  • Relative Amplitude Changes: Relative changes in F1 amplitude between vowel and fricative segments varied significantly, with decreases between Phone1 and Phone2 (vowel to fricative) and increases between Phone2 and Phone3 (fricative to vowel).

Limitations and Future Work

  • Limitations of Previous Methods: The paper highlights the limitations of existing voicing classification methods, particularly the difficulty in interpreting subtle EGG signals and the reliance on subjective interpretation.
  • Future Work: The authors plan to:
  • Define the laryngeal mechanisms and vibration modes associated with weak voicing in normal speakers.
  • Draw parallels between normal weak voicing and conditions in UVFP patients.
  • Develop a model of UVFP voicing to enhance voice rehabilitation therapy.
  • Further investigate the relationship between the new "weak voicing" criterion and EGG signals.

Summary updated

BibTeX
@inproceedings{pinho2009aerodynamics,
  author = {Cátia M. R. Pinho and Luis M. T. Jesus and Anna Barney},
  title = {{Aerodynamics of fricative production in european portuguese}},
  booktitle = {Interspeech},
  year = {2009},
}
Figure