Ultra-compact transformer architecture for real-time RPM estimation from STFT spectrograms. Achieves R²=0.883, MAE=104.09 RPM with 3.5MB model size. Includes pre-trained model on Hugging Face.
machine-learning real-time deep-learning signal-processing pytorch embedded-systems transformer spectrogram stft automotive attention-mechanism frequency-domain audio-processing nvh automotive-engineering ultra-compact rpm-estimation engine-monitoring
-
Updated
Aug 6, 2025 - Python