Acoustic Analysis of Detailed Three-Dimensional Shape of the Human Nasal Cavity and Paranasal Sinuses

Tatsuya Kitamura, Hironori Takemoto, Hisanori Makinae, Tetsutaro Yamaguchi, Kotaro MakiInterspeech 2017

TL;DR

This paper bridges the gap between detailed anatomical models of the nasal cavity and paranasal sinuses and their acoustic impact on speech. By creating a physical 3D model from CT scans and validating it with numerical simulations, the researchers found that these cavities not only contribute to spectral dips (anti-resonances) but also to spectral peaks (resonances), challenging traditional views of their function.

What Problem Does This Paper Address?

The human nasal cavity and paranasal sinuses possess intricate, asymmetric shapes that are known to influence speech acoustics and contribute to speaker individuality. However, previous research has been limited by insufficient resolution in imaging techniques (like MRI) and the inadequacy of simplified acoustic models (like 1D models) to capture the complex interplay of these structures. This has left a gap in understanding the precise acoustic contributions of these cavities, particularly the role of the paranasal sinuses, which have traditionally been viewed primarily as passive side branches. This paper aims to provide a more accurate acoustic analysis by leveraging high-resolution 3D data and advanced modeling techniques.

Methods

The researchers employed a multi-pronged approach:

  • 3D Reconstruction: High-resolution volumetric X-ray CT data from a single female subject was used to reconstruct the 3D geometry of the nasal cavity and paranasal sinuses. This involved median filtering to reduce noise and a supervised machine learning approach (Random Forest classifier) for accurate segmentation of air and tissue regions.
  • Physical Model Creation: The segmented 3D data was then used to create a physical model of the nasal and paranasal cavities using a 3D printer with a layer thickness of 0.254 mm. This model was designed to allow acoustic input at the glottis and measurement at the nostrils.
  • Acoustic Measurement: The transfer function of the physical model was measured in a soundproof room. White noise was introduced at the glottis, and the acoustic response was recorded 20 mm from the nostrils. The measurement setup included compensation for the transfer function of the entire system using inverse filtering.
  • Numerical Simulation (FDTD): The same 3D geometry was used for Finite-Difference Time-Domain (FDTD) simulations to calculate the transfer function. This method allows for detailed acoustic wave propagation modeling within the complex geometry.
  • Sinus Isolation: To understand the individual contribution of each paranasal sinus, simulations were re-run with each sinus (maxillary, sphenoidal, frontal) progressively removed from the model.

Datasets

The study utilized volumetric X-ray CT data of the head and neck regions of an 18-year-old Japanese female patient with no abnormalities of the nasal and paranasal cavities. This data was acquired using a CB MercuRay dentomaxillofacial cone-beam X-ray CT system. The paper does not introduce a new benchmark dataset but rather uses existing CT data to create a novel physical and computational model.

Key Insights and Ideas

  • Validation of 3D Modeling: The study successfully demonstrated that combining high-resolution 3D CT data, 3D printing, and FDTD simulations can accurately capture the acoustic properties of the complex nasal and paranasal cavities.
  • Challenging Traditional Views: The most significant insight is that paranasal sinuses appear to contribute not only to spectral dips (anti-resonances, as traditionally thought) but also to spectral peaks (resonances). This suggests they act as more than just passive side branches.
  • Individual Sinus Contributions: The systematic removal of individual sinuses revealed their specific roles in shaping the transfer function, with some sinuses contributing to spectral dips while others were implicated in generating spectral peaks.
  • Methodological Synergy: The close agreement between the physical model measurements and FDTD simulations validates both the experimental setup and the numerical approach for analyzing complex acoustic geometries.

Key Results

  • The measured and calculated transfer functions of the physical model and the FDTD simulation showed similar gross shapes and frequencies of spectral peaks and dips, indicating the reliability of both methods.
  • Eliminating the left maxillary sinus (LMS), right maxillary sinus (RMS), and sphenoidal sinus (SS) individually resulted in the disappearance of spectral dips at specific frequencies (0.45 kHz, 0.35 kHz, and 0.7 kHz, respectively), consistent with previous findings.
  • However, the elimination of sinuses also led to the disappearance of spectral peaks and the appearance of peak-dip pairs, suggesting that these sinuses actively contribute to generating peaks in the transfer function. For instance, eliminating the LMS resulted in the disappearance of spectral peaks at 1.55 kHz and 3.95 kHz.

Limitations and Future Work

  • Single Subject: The study was based on CT data from a single female subject. The authors acknowledge that additional subjects are needed to confirm their findings and explore individual variations.
  • Wall Vibration Ignored: The FDTD method used in this study, like previous work, ignored sound transmission through the thin walls of the paranasal sinuses. The authors suggest that this wall vibration could be a factor contributing to the observed discrepancies between measurement and simulation and warrants further investigation.
  • 3D Printer Accuracy: The physical model was created using a consumer-grade 3D printer, which may have limitations in accuracy compared to industrial stereolithographs used in prior studies. This could also contribute to discrepancies.
  • Future Directions: The authors propose that further research into the acoustic properties of detailed nasal and paranasal cavities, potentially incorporating wall vibration and a wider range of subjects, could lead to a new perspective on the acoustic theory of speech production.

Summary updated

BibTeX
@inproceedings{kitamura2017acoustic,
  author = {Tatsuya Kitamura and Hironori Takemoto and Hisanori Makinae and Tetsutaro Yamaguchi and Kotaro Maki},
  title = {{Acoustic Analysis of Detailed Three-Dimensional Shape of the Human Nasal Cavity and Paranasal Sinuses}},
  booktitle = {Interspeech},
  year = {2017},
}
Figure