Higher-Order Meta-AF

Meta-Learning for Adaptive Filters with Higher-Order Frequency Dependencies


Adaptive filters are applicable to many signal processing tasks including acoustic echo cancellation, beamforming, and more. Adaptive filters are typically controlled using algorithms such as least-mean squares(LMS), recursive least squares(RLS), or Kalman filter updates. Such models are often applied in the frequency domain, assume frequency independent processing, and do not exploit higher-order frequency dependencies, for simplicity. Recent work on meta-adaptive filters, however, has shown that we can control filter adaptation using neural networks without manual derivation, motivating new work to exploit such information. In this work, we present higher-order meta-adaptive filters, a key improvement to meta-adaptive filters that incorporates higher-order frequency dependencies. We demonstrate our approach on acoustic echo cancellation and develop a family of filters that yield multi-dB improvements over competitive baselines, and are at least an order-of-magnitude less complex. Moreover, we show our improvements hold with or without a downstream speech enhancer.

For more details, please see: “Meta-Learning for Adaptive Filters with Higher-Order Frequency Dependencies”, Junkai Wu, Jonah Casebeer, Nicholas J. Bryan, and Paris Smaragdis, arXiv, 2022. If you use ideas or code from this work, please cite our paper:

  title={Meta-Learning for Adaptive Filters with Higher-Order Frequency Dependencies},
  author={Wu, Junkai and Casebeer, Jonah and Bryan, Nicholas J. and Smaragdis, Paris},    
  booktitle={IEEE International Workshop on Acoustic Signal Enhancement (IWAENC)},

We release code and model checkpoints using the metaaf python package developed for this work. For demos of the code, setup instructions, and more, check out the GitHub repo. To listen to model outputs, keep scrolling.

We release model inputs and outputs for both the linear acoustic echo canceller (AEC) as well as the deep neural network noise suppressor (DNN-NS) post-processor. These results are for all models shown in figure 4 of the paper. When applicable, the top audio file is the output of the linear AEC and the bottom is the output of the DNN-NS post filter. All files contain double-talk and may contain nonlinearities and near/far -end noise.

Demo Files

Near-end Far-end Near-end Speech NLMS RLS Kalman Filter Only DNN-NS Diag. Meta-AF Banded-9 Meta-AF Banded-3 Meta-AF

These samples are all generated using the Meta-AF codebase in this GitHub repo. The code and checkpoints have been released as an add-on to the Meta-AF repository. You may also be interested in checking out the initial Meta-AF video, pre-print, code, demos, and website.