Physics-Informed Neural Operator for Speech Production Analysis

2026-06-21

AI Summary

This work proposes a novel approach to speech production modeling by introducing the Physics-Informed Neural Operator (PINO) for the first time, circumventing the need for time-consuming numerical simulations traditionally used in inverse problem solving and real-time vocal tract simulation. Built upon the one-dimensional acoustic wave equation, the method takes vocal tract geometry as input and end-to-end predicts fundamental frequency, glottal volume velocity, and lip pressure without requiring precomputed training data or iterative solvers. Leveraging GPU parallelization, the framework achieves high computational efficiency while preserving physical consistency. In evaluations across five static vowels, it yields remarkably low errors—0.8% in glottal volume velocity and 3.2% in synthesized waveform—significantly outperforming conventional Runge–Kutta and finite-difference methods.

Abstract

Physics-informed neural operators (PINOs) have recently gained attention as fast numerical simulators with potential for solving inverse problems. This study proposes the first PINO-based method for speech production analysis. The model learns the governing one-dimensional wave equations directly without requiring pre-computed supervised training data. Using vocal tract shape data as input features, we compare the proposed model's predicted f0, glottal volume velocity and sound pressure at the lip for five static vowels to a conventional Runge Kutta/Finite difference approach. With errors of 0.8% for glottal volume flow and 3.2% for speech waveforms, the proposed model enables efficient GPU-parallelized simulation without iterative calculations. We conclude that PINO is a promising approach for fast analysis of speech.

Problem

• speech production analysis
• physics-informed neural operator
• numerical simulation
• inverse problems
• wave equations

Innovation

• Physics-Informed Neural Operator
• Speech Production Modeling
• Wave Equation Learning
• GPU-Parallelized Simulation
• Inverse Problem in Speech