Few-shot sound recognition using attentional similarity

Pytorch implementation of [Learning to match transient sound events using attentional similarity for few-shot sound recognition] (paper)

Citation

If you use this code in your research, please cite our paper.

@inproceedings{chou2019learning,
    title={Learning to match transient sound events using attentional similarity for few-shot sound recognition},
    author={Szu-Yu Chou and Kai-Hsiang Cheng and Jyh-Shing Roger Jang and Yi-Hsuan Yang},
    booktitle = {Proceedings of the International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
    year={2019}
}

Requirements

• Python 2.7
• PyTorch 0.4.0
• LibROSA 0.5.0
• Cuda-9.0

Getting Started

1. Mel-spec data of (noise) ESC50 (link)

We provide the mel-spectrogram data, which extracted from wave files with default parameters showed in main.py. Once the data acquired, please unzip data.zip and have data under attentional-similarity folder.

data contains following files :

• ESC_sep.npy: mel-spec of ESC50
• ESC_noise_sep.npy: mel-spec of noise ESC50
• ESC_tag.npy: class index of ESC_sep and ESC_noise_sep entries
• ESC_tag2idx.npy: matching of class names and class indices

2. Train

2.1 Specify cuda device

It's an optional choice to modify CUDA_VISIBLE_DEVICES in Trainer.py to specify the gpu device that the model is going to run on. The default setting is 0

import os
os.environ['CUDA_VISIBLE_DEVICES'] = '0'

2.2 Start training

Run main.py script and specify the dataset name

• ESC50

    $ python main.py --dn ESC_50
  

• noise ESC50

    $ python main.py --dn ESC_noise_50
  

3. Test with trained model

3.1 Specify cuda device

It's an optional choice to modify CUDA_VISIBLE_DEVICES in Tester.py to specify the gpu device that the model is going to run on. The default setting is 0

import os
os.environ['CUDA_VISIBLE_DEVICES'] = '0'

3.2 Specify trained model path

Set model path pmp in main_test.py to the path that store the model you are going to test.

3.3 Start testing

Run main_test.py script and specify the dataset name

• ESC50

    $ python main_test.py --dn ESC_50
  

• noise ESC50

    $ python main_test.py --dn ESC_noise_50
  

License

This project is licensed under the MIT License - see the LICENSE.md file for details