Efficient neural networks for real-time modeling of analog dynamic range compression

Steinmetz CJ, Reiss JD. Efficient neural networks for real-time modeling of analog dynamic range compression. In: AES Convention 152. Audio Engineering Society; 2022. Paper 10596. Available from: https://aes.org/publications/elibrary-page/?id=21709

@inproceedings{Steinmetz2022_21709,
  author = {Steinmetz, Christian J. and Reiss, Joshua D.},
  title = {{Efficient neural networks for real-time modeling of analog dynamic range compression}},
  booktitle = {AES Convention 152},
  note = {Paper 10596},
  year = {2022},
  month = may,
  publisher = {Audio Engineering Society},
  url = {https://aes.org/publications/elibrary-page/?id=21709}
}

TY  - CPAPER
TI  - Efficient neural networks for real-time modeling of analog dynamic range compression
AU  - Steinmetz, Christian J.
AU  - Reiss, Joshua D.
T2  - AES Convention 152
M1  - Paper 10596
PY  - 2022
DA  - 2022/05/06
UR  - https://aes.org/publications/elibrary-page/?id=21709
PB  - Audio Engineering Society
LA  - en
AB  - Deep learning approaches have demonstrated success in modeling analog audio effects. Nevertheless, challenges remain in modeling more complex effects that involve time-varying nonlinear elements, such as dynamic range compressors. Existing neural network approaches for modeling compression either ignore the device parameters, do not attain sufficient accuracy, or otherwise require large noncausal models prohibiting real-time operation. In this work, we propose a modification to temporal convolutional networks (TCNs) enabling greater efficiency without sacrificing performance. By utilizing very sparse convolutional kernels through rapidly growing dilations, our model attains a significant receptive field using fewer layers, reducing computation. Through a detailed evaluation we demonstrate our efficient and causal approach achieves state-of-the-art performance in modeling the analog LA-2A, is capable of real-time operation on CPU, and only requires 10 minutes of training data.
ER  -