Sliding vocal-tract model and its application for vowel production

Takayuki AraiInterspeech 2009

TL;DR

This paper explores a "sliding vocal-tract model" that uses a movable inner cylinder to simulate tongue constriction, allowing for the production of various vowel sounds. The authors demonstrate its ability to cover a significant portion of the vowel space and discuss its potential applications in speech education and research.

What Problem Does This Paper Address?

Understanding how the human vocal tract produces different vowel sounds is a fundamental challenge in acoustics and speech science. While physical models have been developed to illustrate these concepts, many are static or overly complex. This paper addresses the need for a simplified, yet effective, physical model that can dynamically demonstrate the relationship between vocal tract shape and vowel acoustics, particularly focusing on the role of tongue constriction. The authors aim to show how a single degree of freedom (constriction position) can significantly alter vowel quality and explore how additional parameters can expand the model's capabilities for educational and research purposes.

Methods

The core of this work is the "sliding three-tube (S3T) model," an evolution of Fant's three-tube model. This model consists of a long outer cylinder representing the main vocal tract and a shorter inner cylinder that can slide along its length. This sliding inner cylinder simulates the tongue's constriction, a key factor in vowel articulation.

The authors employed two primary methods:

  • Simulation: They used a speech production simulator (Vtcalcs) to model the acoustic behavior of the S3T model. This involved varying the diameter (d) and length (l2) of the inner cylinder and its position within the outer cylinder (L) to generate simulated formant frequencies (F1 and F2).
  • Physical Measurement: They constructed a physical S3T model using acrylic resin tubes. This model was driven by an impulse train to simulate the glottal source, and the resulting output sounds were recorded and analyzed for formant frequencies using linear prediction.

They also investigated modifications to the model, including:

  • Laryngeal Constriction: Adding a narrower section at the glottis end to simulate laryngeal articulation.
  • Lip Rounding: While not physically implemented in the described model, the authors discuss its importance and how it would affect the acoustic output.
  • Variable Constriction Area: Designing a slider mechanism that allows for continuous adjustment of the constriction's cross-sectional area, in addition to its position.

Datasets

The paper does not introduce new datasets for training or evaluation. Instead, it relies on:

  • Existing measurements: The authors refer to and compare their results with previous measurements of the S3T model [1] and data from Chiba & Kajiyama [5] and Stevens [6] for American English vowels.
  • Simulated data: The primary "dataset" is generated through the Vtcalcs simulator, where formant frequencies are computed based on the model's geometric parameters.

Key Insights and Ideas

  • Simplicity and Power of a Single Degree of Freedom: The S3T model effectively demonstrates that by simply sliding the inner cylinder (changing the constriction's position), a wide range of vowel sounds can be produced. This highlights the significant acoustic impact of tongue movement.
  • Vowel Space Coverage: The model, even in its basic form, can cover a substantial portion of the F1-F2 vowel space, particularly the extreme vowels like /i/, /u/, /o/, and /a/.
  • Impact of Constriction Diameter: A smaller constriction diameter (d) leads to a wider spread of vowels in the F1-F2 plot, while a larger diameter compresses the vowel space towards the center.
  • Laryngeal Constriction Enhances High-Front Vowels: Adding a laryngeal constriction improves the model's ability to produce high-front vowels like /i/.
  • Lip Rounding for Low Vowels: The authors conceptually link lip rounding (and protrusion) to the production of low vowels like /u/, suggesting it's a crucial factor for extending the model's range.
  • Educational Value: The model's intuitive design makes it an excellent tool for teaching fundamental concepts of speech production, acoustics, and the source-filter theory to students and even children.

Key Results

  • Vowel Trajectories: Simulations and physical measurements show similar F1-F2 trajectories as the constriction position is varied. These trajectories generally move from high-front vowels (/i/) towards low vowels (/a/ or /ae/) and then back towards high-back vowels (/u/).
  • Constriction Diameter Effect: Increasing the constriction diameter from 10 mm to 20 mm resulted in a compression of the vowel space trajectories towards the center.
  • Laryngeal Constriction Improvement: The addition of a laryngeal constriction (20 mm long, 10 mm diameter) was shown to improve the model's representation of the /i/ vowel.
  • Formant Frequency Approximation: The simple calculations for resonance frequencies of the three tube segments provided a good approximation to the measured and simulated formant frequencies.

Limitations and Future Work

  • Limited Vowel Range: While the model covers a good portion of the vowel space, it doesn't fully capture all nuances, particularly for mid-vowels. The authors suggest that adding more degrees of freedom, such as lip rounding and more complex constriction shapes, could improve coverage.
  • Simplification of Vocal Tract: The model is a simplification of the complex, three-dimensional human vocal tract. The authors acknowledge that the tube is not perfectly uniform and that the glottal source is also simplified.
  • Future Work: The authors suggest further development of the model to include lip rounding and more sophisticated constriction mechanisms. They also propose its use in investigating how the auditory system segregates vocal tract shape from speaker size information. The head-shaped version of the model is also highlighted as a useful educational tool.

Summary updated

BibTeX
@inproceedings{arai2009sliding,
  author = {Takayuki Arai},
  title = {{Sliding vocal-tract model and its application for vowel production}},
  booktitle = {Interspeech},
  year = {2009},
}
Figure