ERANNs

An implementation of ERANNs (Efficient Residual Audio Neural Networks for Audio Pattern Recognition), built using the Keras API from TensorFlow.

This architecture is based on the article "ERANNs: Efficient residual audio neural networks for audio pattern recognition", published in Pattern Recognition Letters.

I sincerely thank and acknowledge the authors of the original work — Sergey Verbitskiy, Vladimir Berikov, and Viacheslav Vyshegorodtsev — for their valuable contribution to the field. This implementation closely follows the architectural details presented in their publication.

---

📦 Installation

Clone this repository and install the package locally using pip:

git clone https://github.com/Matheuao/ERANNs.git
cd ERANNS
pip install .

Requirements: Python ≥ 3.6, TensorFlow ≥ 2.4.0

If you're using a platform like Kaggle Notebooks, you can also install directly within a cell:

!git clone https://github.com/Matheuao/ERANNs.git
!pip install ./ERANNS

---

🚀 Usage

Once installed, you can import and instantiate the ERANNs model as follows:

from eranns import ERANNs

# Create an ERANNs model with sm=2, W=4, and 527 output classes
model = ERANNs(sm=2, W=4, T0=128, N=527)
model.summary()

---

📚 Parameters

The ERANNs() function accepts the following parameters:

Parameter	Description
sm	Stride mode that controls temporal/frequency resolution
W	Base width multiplier for the number of filters in each stage
T0	Temporal input dimension (default: 128)
N	Number of output classes (default: 527)

---

📁 Project Structure

ERANNs/
├── eranns/
│   ├── __init__.py              # Main interface
│   ├── layers.py                # ARB block implementation
│   └── model.py                 # ERANNs architecture definition
├───├── data/
|       ├──__init__.py           # Main interface
|       ├──esc_50.y              # Train dataset pipeline
|       └──esc_50_augmented.py   # Train dataset pipeline with data augmentation(still in development)
├── setup.py                     # Package setup script
├── requirements.txt             # Dependencies
└── README.md                    # This file

---

📜 License

This project is licensed under the MIT License.

---

🤝 Contributing

Contributions are welcome! If you have suggestions or encounter issues, feel free to open a GitHub Issue or submit a pull request.