Codebase for "Transcription free filler word detection with Neural semi-CRFs" [ICASSP2023], arxiv link:
This repository requires Python 3.X and Pytorch 1.X. Other packages are listed in requirements.txt.
Reference: Structured State Spaces for Sequence Modeling
Cauchy Kernel A core operation of S4 is the "Cauchy kernel" described in the paper. This is actually a very simple operation; a naive implementation of this operation can be found in the standalone in the function cauchy_naive. However, as the paper describes, this has suboptimal memory usage that currently requires a custom kernel to overcome in PyTorch.
Two more efficient methods are supported. The code will automatically detect if either of these is installed and call the appropriate kernel.
Custom CUDA Kernel This version is faster but requires manual compilation for each machine environment. Run python setup.py install from the directory extensions/cauchy/.
Edit the configuration file in params.py to specify the model hyperparameters and dataset paths. Then run train.py to train the model.
In this implementation, we directly use the VAD predictions provided by the PodcastFillers Dataset, which is based on GPVAD, these predictions could be replaced by any other pretrained VAD models.
@article{yan2021skipping,
title={Skipping the frame-level: Event-based piano transcription with neural semi-crfs},
author={Yan, Yujia and Cwitkowitz, Frank and Duan, Zhiyao},
journal={Advances in Neural Information Processing Systems},
volume={34},
pages={20583--20595},
year={2021}
}@inproceedings{gu2022efficiently,
title={Efficiently Modeling Long Sequences with Structured State Spaces},
author={Gu, Albert and Goel, Karan and R\'e, Christopher},
booktitle={The International Conference on Learning Representations ({ICLR})},
year={2022}
}@article{zhu2022filler,
title={Filler Word Detection and Classification: A Dataset and Benchmark},
author={Zhu, Ge and Caceres, Juan-Pablo and Salamon, Justin},
journal={arXiv preprint arXiv:2203.15135},
year={2022}
}