Gradient-Based Learning of Parametric Engine Sound Representations for Real-Time Resynthesis and Tuning on Embedded Systems

Doerfler R, Kuntz M, Zimmer C. Gradient-Based Learning of Parametric Engine Sound Representations for Real-Time Resynthesis and Tuning on Embedded Systems. In: 2026 AES 6th International Automotive Audio Conference. Audio Engineering Society; 2026.pp. 27:1-11. Available from: https://aes.org/publications/elibrary-page/?id=23420

@inproceedings{Doerfler2026_23420,
  author = {Doerfler, Robin and Kuntz, Matthieu and Zimmer, Clemens},
  title = {{Gradient-Based Learning of Parametric Engine Sound Representations for Real-Time Resynthesis and Tuning on Embedded Systems}},
  booktitle = {2026 AES 6th International Automotive Audio Conference},
  pages = {27:1-11},
  url = {https://aes.org/publications/elibrary-page/?id=23420},
  year = {2026},
  month = jul,
  publisher = {Audio Engineering Society}
}

TY  - CPAPER
TI  - Gradient-Based Learning of Parametric Engine Sound Representations for Real-Time Resynthesis and Tuning on Embedded Systems
AU  - Doerfler, Robin
AU  - Kuntz, Matthieu
AU  - Zimmer, Clemens
T2  - 2026 AES 6th International Automotive Audio Conference
SP  - 1
EP  - 11
M1  - Paper 27
PY  - 2026
DA  - 2026/07/29
UR  - https://aes.org/publications/elibrary-page/?id=23420
PB  - Audio Engineering Society
LA  - en
AB  - Engine order enhancement is central in automotive sound design, where selective harmonics are synthesized to shape perceptual qualities such as sportiness, refinedness, or power. This paper investigates a neural network-based approach to combustion engine sound modeling that extends conventional engine order analysis and enhancement by deriving synthesis parameters from audio data with machine learning and incorporating stochastic components into the synthesis framework. The system parameterizes engine sounds as a compact representation capturing per-order and broadband timbral variation across the full RPM-torque operating range, while remaining manually tunable and compatible with established automotive audio frameworks. The approach leverages gradient-based optimization and analysis-by-synthesis through an end-to-end differentiable implementation. The resulting synthesis parameter set is directly transferable to conventional DSP implementations for deployment on embedded targets. Spectral metrics and listening tests1confirm high reconstruction fidelity, and integration into an established automotive audio development platform (EVx Suite) demonstrates technical feasibility on deployment-ready embedded systems.
ER  -