Code for Taejun Kim and Juhan Nam, "Temporal Feedback Convolutional Recurrent Neural Networks for Speech Command Recognition," APSIPA ASC, 2022 [pdf]
This repository is tested under Python 3.10.
curl -O http://download.tensorflow.org/data/speech_commands_v0.02.tar.gz
mkdir dataset
tar zxvf speech_commands_v0.02.tar.gz -C ./datasetInstall PyTorch according to your environment at the official website, and run:
pip install -e .It will install tfcrnn package and its dependencies.
Weights & Biases (W&B) is integrated so you can use its nice visualizations if you sign up
and log in to W&B using wandb login. Though, you can also run the code without an account.
By default, it will train a TF-CRNN with the basic block:
python tfcrnn/train.pyIf you want to train another type of network, use --skeleton cnn|crnn|tfcrnn and --block basic|se|resse:
# An example for training a SampleCNN with Res-SE blocks.
python tfcrnn/train.py --skeleton cnn --block resse@inproceedings{taejun2022tfcrnn,
title={Temporal Feedback Convolutional Recurrent Neural Networks for Speech Command Recognition},
author={Kim, Taejun and Nam, Juhan},
booktitle={Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC)},
year={2022},
organization={IEEE}
}