TOWARD REAL-TIME PHYSICALLY-BASED VOICE SIMULATION: AN EIGENMODE-BASED APPROACH

Area of Science:

• Computational physics
• Bioacoustics
• Speech science

Background:

• Physically-based continuum models of voice production are valuable for clinical voice disorder intervention and personalized speech synthesis.
• High computational cost of fluid-structure interaction in current models limits their practical application.
• Developing computationally efficient models is crucial for advancing voice production research.

Purpose of the Study:

• To present a computationally efficient, physically-based continuum model for voice production.
• To enable near real-time applications in voice analysis and synthesis.
• To reduce the computational burden of simulating voice production.

Main Methods:

• Developed an eigenmode-based formulation for the governing equations of voice production.
• Utilized vocal fold eigenmodes as fundamental components to reconstruct complex vibration patterns.
• Reduced the degrees of freedom compared to traditional finite element models.

Main Results:

• Simulations achieved reasonable accuracy in fundamental frequency, vocal intensity, and spectral measures using approximately 100 vocal fold eigenmodes.
• Significantly reduced computational time and degrees of freedom compared to finite element models.
• Demonstrated the model's capability to simulate a wide range of voice qualities and natural voice changes.

Conclusions:

• The eigenmode-based continuum model offers a computationally efficient alternative for voice production simulation.
• The model shows promise for real-time applications in voice disorder treatment and speech synthesis.
• Further reduction in eigenmode count may be acceptable for applications where absolute precision is not critical.