Problem

Classifying songs into different genres is a time-consuming task that requires human effort. This can be a problem for music streaming services that need to categorize their vast libraries of songs. It can also be a challenge for music enthusiasts who want to organize their personal collections. The existing solutions are either inaccurate or require manual input, which is not practical.

To learn more about audio processing, I decided to work on this project wich consists of creating an automated music genre classifier that could accurately categorize songs into 10 different genres without any human intervention.

Check out the complete demo on my github: https://github.com/Abdelkabir-menani/music_genre_classifier

Solution

We developed a convolutional neural network (CNN) model with three convolutional layers and a learning rate of 0.0001. We used the GTZAN dataset, which contains 1000 audio files of 30 seconds each, evenly distributed across 10 genres. To extract features from the audio files, we used Mel Frequency Cepstral Coefficients (MFCCs), which are commonly used in speech recognition and music information retrieval.

We built a web application using FastAPI and MongoDB to deploy our model. Users can upload an audio file to the app, and it will return the predicted genre of the song within seconds. We also containerized our app using Docker for easy deployment.

Impact

Achieved an accuracy rate of 78% on classifying songs into their respective genres

skills:

docker - fast api - CNN - audio processing - Mel Frequency Cepstral Coefficients

Automated Music Genre Classifier with CNN