This project, developed over a semester as part of the Apply AI group at AI4All, involved collaborative efforts with two peers and guidance from an AI mentor. Our objective was to utilize Convolutional Neural Networks (CNN) and Computer Vision techniques to accurately identify specific birdcalls from audio data. This initiative aimed to contribute to the field of ornithology by providing an efficient tool for bird species identification through their calls.
- Convolutional Neural Networks (CNNs): Leveraged advanced CNN architectures to process and classify audio spectrograms of birdcalls.
- Computer Vision: Implemented image processing techniques to convert audio signals into visual representations (spectrograms) for better analysis and identification.
- Data Augmentation: Employed data augmentation strategies to enhance the training dataset, ensuring robust model performance.
- Accuracy and Precision: Achieved high levels of accuracy in identifying bird species, demonstrating the effectiveness of the applied methodologies.
Python: Core programming language for developing the application. PyTorch: Utilized for building and training the CNN models. Torchaudio: Used for audio processing and generating spectrograms. NumPy and Pandas: Employed for data manipulation and analysis. Matplotlib: Utilized for visualizing the spectrograms and model performance metrics.
The project is currently unfinished, with the following components under development:
- Dataset Preparation: Additional data collection and preprocessing, including handling truncated and corrupt files.
- Model Training: Further refinement of the CNN model, including hyperparameter tuning and extended training epochs.
- Evaluation Metrics: Implementation of more comprehensive evaluation metrics to assess model performance.
- User Interface: Development of a user-friendly interface for easy interaction with the model.
- Deployment: Setting up the model for deployment in a production environment, including potential integration with web services or mobile applications.
We welcome contributions from the community. Please fork the repository and submit a pull request with detailed information about the changes. For significant modifications, open an issue to discuss your ideas before implementation.
We extend our gratitude to AI4All and our AI mentor for their invaluable support and guidance throughout this project. Their expertise and encouragement were instrumental in our success.