Classify Sound from Audio Data

Inspiration

The idea for Classify Song Genres from Audio Data was inspired by my passion for music and machine learning. I wanted to explore how technology can be used to understand and categorize the diverse world of music.

What I Learned

Through this project, I gained insights into audio signal processing, feature extraction, and the intricacies of training machine learning models on audio data. I also learned how to handle large datasets and the importance of model evaluation and tuning.

How I Built the Project

• Data Collection: Gathered a diverse dataset of songs labeled by genre.
• Feature Extraction: Utilized libraries like LibROSA to extract audio features such as Mel-frequency cepstral coefficients (MFCCs).
• Model Development: Implemented a Convolutional Neural Network (CNN) using TensorFlow/Keras to classify the genres.
• Training and Evaluation: Trained the model on a GPU for efficiency and evaluated its performance using metrics like accuracy and confusion matrix.

Challenges Faced

One of the major challenges was preprocessing the audio data to ensure consistent and meaningful feature extraction. Additionally, tuning the CNN model to achieve high accuracy without overfitting required careful experimentation and cross-validation.

Built With

• librosa
• numpy
• python
• tensorflow