Deep Learning Techniques for Music Genre Classification and Building a Music Recommendation System

Recommendation mechanisms have been increasingly popular in recent years when a large number of people rely on the internet to discover solutions from a wide variety of choices. Due to the competition of the music market, companies are committed to providing personalized music to users in order to attract more buyers. Recommending music that takes into account the music features can enhance the user’s listening experience and increase consumer service. In this study, a music recommendation system is built after classifying the tracks according to the genre. The convolutional neural network (CNN), convolutional neural network with long short-term memory (CNN-LSTM), and convolutional neural network with bidirectional long short-term memory (CNN-BiLSTM) models are used for the classification. CNN is considered as the base model. The CNN-LSTM and CNN-BiLSTM models are built on it. The data is taken from the free music archive (FMA) dataset. The content-based (CB) recommendation system with cosine similarity is used as a recommendation model. The features are extracted using a Mel-spectrogram from the audio files. On evaluation, it was observed that the CNN-LSTM model with CB recommendations performed the best. In the future, an ensembled model consisting of all 3 classification models to classify the genre along with a hybrid CB-CF system can be used