Brief

Researchers have developed a novel neural synthesis model called the Pulse-Train-Resonator (PTR) for generating realistic engine sounds. Unlike previous methods that focus on spectral characteristics, PTR directly models the underlying pulse shapes and temporal structure of engine audio. The model integrates physics-informed biases, such as harmonic decay and thermodynamic pitch modulation, to simulate exhaust acoustics and engine firing patterns. In evaluations, PTR demonstrated a significant improvement in harmonic reconstruction and a reduction in overall loss compared to existing baseline models. AI

Researchers have developed a novel framework for procedurally generating engine sounds with precise control annotations, addressing the difficulty of obtaining such data for automotive audio applications. This method extracts harmonic structures from real recordings to drive a parametric synthesizer, augmenting limited source audio to create a large dataset of engine sounds with sample-accurate RPM and torque annotations. The resulting dataset, comprising 19 hours of audio, has been validated against real recordings and demonstrated its utility in training a synthesis network, making it publicly available to advance research in engine sound analysis and modeling. AI