Improving Low-Latency Automatic Drum Transcription for Automotive Applications

Weyers P, Uhle C, Mueller M, Lang M. Improving Low-Latency Automatic Drum Transcription for Automotive Applications. In: 2026 AES 6th International Automotive Audio Conference. Audio Engineering Society; 2026. Paper 8. Available from: https://aes.org/publications/elibrary-page/?id=23402

@inproceedings{Weyers2026_23402,
  author = {Weyers, Philipp and Uhle, Christian and Mueller, Meinard and Lang, Matthias},
  title = {{Improving Low-Latency Automatic Drum Transcription for Automotive Applications}},
  booktitle = {2026 AES 6th International Automotive Audio Conference},
  note = {Paper 8},
  year = {2026},
  month = jul,
  publisher = {Audio Engineering Society},
  url = {https://aes.org/publications/elibrary-page/?id=23402}
}

TY  - CPAPER
TI  - Improving Low-Latency Automatic Drum Transcription for Automotive Applications
AU  - Weyers, Philipp
AU  - Uhle, Christian
AU  - Mueller, Meinard
AU  - Lang, Matthias
T2  - 2026 AES 6th International Automotive Audio Conference
M1  - Paper 8
PY  - 2026
DA  - 2026/07/29
UR  - https://aes.org/publications/elibrary-page/?id=23402
PB  - Audio Engineering Society
LA  - en
AB  - In automotive audio systems, musical beats and drum events can control synchronized in-cabin experiences such as ambient lighting and music-driven visual effects. Compared to beat tracking, automatic drum transcription (ADT) offers richer control signals by detecting and classifying drum onsets of multiple drum classes (e.g., bass drum, snare drum, and hi-hat), enabling more precise and musically meaningful synchronization. Deploying ADT in vehicles, however, requires low latency, computational efficiency, and robust performance for various input signals. This paper investigates improvements to low-latency ADT suitable for automotive deployment, using the Separate- Tracks-Annotate-Resynthesize Drums (STAR Drums) dataset and a block-based processing strategy that achieves an average detection delay of around 60 ms. We explore three strategies: (1) lightweight architecture modifications inspired by recent advances in image classification, combined with a temporal convolutional network (TCN); (2) re-rendering STAR Drums to increase drum timbre diversity and augmenting the re-synthesized drum stems; and (3) refinement training with pseudo labels obtained from source-separated mixtures. Our results show that data augmentation and increased drum timbre diversity yield modest performance gains, whereas pseudo-label refinement provides the largest effect, with up to 18% relative improvement in global F-measure. In the real-time eight-class setting, our best model achieves a global F-measure of 0.77 on MDB Drums, competitive with state-ofthe- art offline systems, demonstrating that accurate and efficient ADT is feasible for automotive deployment.
ER  -