Introduction

In the realm of environmental conservation and wildlife research, accurately identifying bird species based on their unique vocalizations is paramount. However, this task presents challenges due to the vast diversity of avian calls and the complexity of acoustic environments. To overcome these challenges, we present a robust solution: Bird Sound Classification using Deep Learning. Leveraging the power of TensorFlow, our project employs Convolutional Neural Networks (CNNs) to analyze audio signals and classify them into distinct bird species with high precision. By harnessing state-of-the-art machine learning techniques, our endeavor seeks to revolutionize avian research, facilitate ecological monitoring, and contribute to the conservation of avifauna worldwide.

Table of Contents

1. Key Technologies and Skills
2. Installation
3. Usage
4. Features
5. Contributing
6. License
7. Contact

Key Technologies and Skills

• Python
• TensorFlow
• Convolutional Neural Network (CNN)
• Keras
• scikit-learn
• OpenCV
• Numpy
• Pandas
• Matplotlib
• Streamlit
• Hugging Face

Installation

To run this project, you need to install the following packages:

pip install tensorflow
pip install scikit-learn
pip install opencv-python
pip install librosa
pip install ipython
pip install numpy
pip install pandas
pip install matplotlib
pip install streamlit
pip install streamlit_extras
pip install tqdm

Note: If you face "ImportError: DLL load failed" error while installing TensorFlow,

pip uninstall tensorflow
pip install tensorflow==2.12.0 --upgrade

Usage

To use this project, follow these steps:

1. Clone the repository: git clone https://github.com/gopiashokan/Bird-Sound-Classification-using-Deep-Learning.git
2. Install the required packages: pip install -r requirements.txt
3. Run the Streamlit app: streamlit run app.py
4. Access the app in your browser at https://localhost:8501

Features

• The bird sound dataset utilized in this project was sourced from Kaggle, a prominent platform for datasets and data science resources. This dataset comprises 2161 audio files (mp3) capturing the vocalizations of 114 distinct bird species.

• Each audio file is meticulously annotated with the corresponding bird species, providing crucial labels for supervised learning tasks in deep learning-based classification models.

📙 Dataset Link: https://www.kaggle.com/datasets/soumendraprasad/sound-of-114-species-of-birds-till-2022

• Audio Feature Extraction: The preprocessing phase commences with the extraction of audio features from the raw audio files using the Librosa library. Specifically, we utilize Librosa to extract Mel Frequency Cepstral Coefficients (MFCC) from each audio signal, capturing crucial spectral characteristics essential for bird sound classification.

• TensorFlow Dataset Creation: Following feature extraction, we convert the extracted MFCC features and corresponding target labels into TensorFlow tensors. These tensors are then encapsulated within a TensorFlow dataset, facilitating seamless integration into the deep learning model architecture.

• Data Splitting: To ensure robust model evaluation, the dataset is partitioned into three subsets: training, validation, and testing. This partitioning scheme enables independent assessment of model performance during training, validation, and final testing stages, thereby enhancing model generalization and mitigating the risk of overfitting.

• Data Pipeline Optimization: A key focus of preprocessing is the optimization of the data pipeline to enhance training efficiency. Leveraging TensorFlow's pipeline optimization techniques, such as caching, shuffling, and prefetching, we accelerate the data ingestion process and minimize training time. By proactively prefetching data batches and caching preprocessed samples, we mitigate potential bottlenecks and maximize GPU utilization, culminating in expedited model convergence and improved computational efficiency.

• Model Architecture: The model architecture is meticulously crafted using TensorFlow's Keras API. A Convolutional Neural Network (CNN) is constructed, featuring convolutional layers for feature extraction, pooling layers for spatial downsampling, and dense layers for classification. Adjustable filters, units, and activation functions are incorporated to tailor the model's capacity to the complexity of the dataset.

• Training: Model training is orchestrated using an end-to-end pipeline encompassing data loading, preprocessing, model instantiation, and optimization. Leveraging the Adam optimizer, sparse_categorical_crossentropy loss function, and Accuracy metrics, we optimize the model parameters to minimize classification error. Throughout training, the model's performance is monitored on a separate validation dataset after each epoch to prevent overfitting and ensure generalization. Upon completion of training, the model attains a remarkable accuracy of 93.4%, underscoring its proficiency in accurately classifying bird sounds.

• Following the completion of model training and evaluation, the trained model is saved to enable seamless deployment and inference on new audio for classification purposes. To facilitate this process, a user-friendly Streamlit application is developed and deployed on the Hugging Face platform.

• This application empowers users to upload new audio files and obtain real-time classification results, providing a convenient interface for leveraging the model's capabilities in practical scenarios.

🚀 Application: https://huggingface.co/spaces/gopiashokan/Bird-Sound-Classification

Contributing

Contributions to this project are welcome! If you encounter any issues or have suggestions for improvements, please feel free to submit a pull request.

License

This project is licensed under the MIT License. Please review the LICENSE file for more details.

Contact

📧 Email: [email protected]

🌐 LinkedIn: linkedin.com/in/gopiashokan

For any further questions or inquiries, feel free to reach out. We are happy to assist you with any queries.