Speech rate effects on european portuguese nasal vowels

Catarina Oliveira, Paula Martins, António TeixeiraInterspeech 2009

TL;DR

This study investigates how speech rate affects the production of European Portuguese nasal vowels using articulatory data. The findings reveal that faster speech leads to shorter velum gestures, increased stiffness, and greater overlap between articulatory movements, offering crucial insights for developing more naturalistic text-to-speech systems.

What Problem Does This Paper Address?

Understanding the temporal dynamics of speech production is crucial for accurately modeling and synthesizing human speech. Nasal vowels, in particular, present a complex articulatory challenge due to the coordinated movements of the velum and oral articulators. While previous research has touched upon the temporal aspects of nasality, there's a need for more detailed information on how factors like speech rate influence the fine-grained timing and dynamics of velum gestures in European Portuguese. This paper aims to fill this gap by providing new temporal information on the production of EP nasal vowels, specifically examining the influence of speech rate on velum gesture duration, stiffness, and inter-gestural coordination. This knowledge is vital for improving articulatory-based text-to-speech (TTS) systems, which rely on realistic gestural representations.

Methods

The researchers employed a two-dimensional electromagnetic transduction device (Carstens AG100) to capture articulatory movements of the velum, upper and lower lips, and tongue tip. Acoustic signals were simultaneously recorded and synchronized with the kinematic data. The data was collected at a sampling rate of 500 Hz for kinematic data and 20 kHz for acoustic signals. Standard calibration and post-processing procedures were applied.

Key to their analysis was the automatic labeling of articulatory events, including movement onset, target achievement, target release, and release offset. These landmarks were identified using velocity thresholds applied to specific articulators (tongue tip for alveolar stops, lip aperture for labial consonants, and velum vertical movement for nasal vowels).

To quantify the dynamics of velum movement, they analyzed:

  • Velum Duration: Measured the total cycle, opening phase, plateau, and closing phase of velum movement.
  • Velum Stiffness: Calculated as the ratio of peak velocity to maximum displacement for both opening and closing phases. This parameter provides a context-independent measure of articulatory movement velocity.
  • Inter-gestural Timing: Assessed the temporal coordination between different gestures, including:
  • TTL (Target-to-Target Lag): The interval between the achievement of the velum opening target and the oral consonant target.
  • T2TL (Target-to-Target Lag): The interval between the achievement of the velum closing target and the oral consonant target.
  • SRL (Start-to-Release Lag): The interval between the onset of velum movement and the release of the preceding oral consonant.
  • Velum-to-Glottis Lag: The interval between the achievement of the velum closing target and the opening of the glottis.

Statistical analysis, primarily using three-way ANOVAs, was performed to evaluate the significance of factors like speech rate, position (initial, internal, final), and nasal vowel type on these measured parameters.

Datasets

The study utilized a newly collected dataset of European Portuguese nasal vowels produced by two native speakers (one male, one female). The corpus included nonwords featuring the five nasal vowels ([ẽ], [ẽ], [ĩ], [ũ], [õ]) in three prosodic conditions (word-initial, medial, final). These were embedded in carrier sentences ("Diz...três vezes") and produced under two speaking rate conditions: normal/comfortable and fast. For this paper, the results from the female speaker are reported. This dataset appears to be a subset of a larger corpus designed for studying Portuguese nasality.

Key Insights and Ideas

  • Speech Rate is a Major Driver of Nasal Vowel Dynamics: The study strongly confirms that speech rate significantly impacts the temporal characteristics of EP nasal vowels. This isn't just about speaking faster; it fundamentally alters how articulators move.
  • Velum Gestures Shrink and Stiffen at Faster Rates: Faster speech leads to shorter durations for velum opening, plateau, and closing phases. Simultaneously, the velum becomes "stiffer," meaning it moves more rapidly and with greater force, which is a direct consequence of the reduced duration.
  • Increased Articulatory Overlap is Key to Speed: To achieve faster speech, articulatory gestures (velum closing, oral consonant, and glottal gestures) overlap more significantly. This increased coordination allows for a more compressed speech stream, potentially leading to the emergence of consonantal segments in faster speech.
  • Portuguese Nasal Vowel Production is Asymmetric: The velum closes more rapidly than it opens, and the opening gesture is generally longer than the closing gesture. This asymmetry is maintained even at faster speech rates.
  • Lexical Position Matters: The position of the nasal vowel within a word (initial, internal, final) also influences timing, though the effects of speech rate are often more pronounced.
  • Implications for TTS: The findings provide concrete, quantifiable data on how articulatory gestures change with speech rate, which is directly applicable to building more natural and dynamic articulatory-based TTS systems.

Key Results

  • Duration: A significant decrease in velum gesture duration (total, opening, plateau, closing) was observed at fast speech rates compared to normal rates. For example, total velum duration decreased from around 200-340 ms to 141-220 ms.
  • Stiffness: Velum stiffness generally increased from normal to fast speech rates across most positions, indicating faster and more forceful movements.
  • Inter-gestural Timing:
    • TTL: Values were consistently negative, indicating the velum opening target precedes the oral consonant target. At fast rates, these negative values became closer to zero, signifying a shorter lag.
  • T2TL: The interval between velum closing target and oral consonant target significantly decreased from normal to fast rates, from around 100-110 ms to 60-70 ms.
  • SRL: At normal rates, velum onset occurred after the release of the preceding labial consonant (positive SRL), but at fast rates, this lag diminished, with gestures becoming nearly simultaneous.
  • Velum-to-Glottis Lag: The glottis opens before the velum fully closes, and this lag was generally smaller at faster speech rates.

Limitations and Future Work

  • Speaker Variability: While the study reports on one female speaker, the authors acknowledge that dialectal and individual factors can affect inter-gestural cohesion. Future work should include a wider range of speakers to capture this variability.
  • Labeling Challenges: The authors note potential labeling issues in the fast speech condition, particularly for the opening phase in final positions, which might explain some unexpected duration differences.
  • Limited Articulators: The study focused on specific articulators. A more comprehensive analysis involving a wider range of articulators could provide a more complete picture.
  • TTS Integration: While the paper highlights the relevance for TTS, further research is needed to directly integrate these findings into articulatory-based TTS systems and evaluate their perceptual impact.
  • Broader Phonetic Context: Investigating these effects across a wider range of phonetic contexts, including different consonant types and vowel qualities, would be beneficial.

Summary updated

BibTeX
@inproceedings{oliveira2009speech,
  author = {Catarina Oliveira and Paula Martins and António Teixeira},
  title = {{Speech rate effects on european portuguese nasal vowels}},
  booktitle = {Interspeech},
  year = {2009},
}
Figure