This repository contains an implementation in Pytorch of a Single Attention Convolutional Neural Network (Single-Att-CNN) model for audio tagging and sound event detection.
The model is trained and evaluated on the ESC-50 dataset, but can be easily modified to be applied on others datasets or used as a base layout for your projects.
The model is comprised of a CNN component that is used as an encoder and a single-attention mechanism that is based on how humans perceive and classify different kinds of sounds.
Clone this repository and install all the dependencies. The usage of a virtualenv or Conda environment is recommended.
git clone https://github.com/alefiury/Audio-Tagging-Single-Attention-CNN
cd Audio-Tagging-Single-Attention-CNN
pip install -r requirements.txt
First, is necessary to download the dataset that you will be using. The ESC-50 dataset can be downloaded with the following bash command:
sudo bash download_dataset.sh
The application uses Hydra to control all the training configurations. For more information on Hydra and it's documentation visit the Hydra website.
The relevant information related with the training configuration can be found in the default.yaml
file, inside the config folder.
You can also pass or modify options through the command line, for instance: python main.py train.epochs=100.
To save training statistics (loss, accuracy, gradients and etc) the Weights & Biases plataform is used. Is necessary that you create an account before running the application. Related information can be found on their website.
To train or evaluate your model you need to set the command option.
- To train:
python main.py command=train - To evaluate:
python main.py command=test
- Alef Iury Siqueira Ferreira
- e-mail: [email protected]
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent