Emotion-based song recommendation for mothers with postpartum depression

Postpartum depression is a mental health disorder that can occur in first-time mothers. Music therapy is one of the treatments to help reduce Postpartum Depression (PPD) symptoms and provide sootheness for postpartum mothers. This study aimed to develop an emotion and genre-based music recommendation system designed to meet the emotional needs of mothers with PPD. A music classification model was developed using cochleogram and chromagram-based audio representations to obtain music emotion and genre features through a Convolutional Recurrent Neural Network (CRNN) architecture. An evaluation was conducted to assess the effect of using both audio representations on music classification performance. The results show that the use of cochleogram and chromagram-based audio representations, which aims to integrate music spectral and key modulation information, can provide better accuracy compared to other music classifications that only utilizes spectral information. The feature extraction results in the classification process are used to make music recommendations based on similarity between the music emotion and genre information. The result of this study is to provide music recommendations that match the emotional state and preferences of mothers with PPD, as well as contribute to the literature related to audio processing for music classification and recommendation. The user evaluation testing was conducted to 35 new mothers and the satisfaction score result is 65%. Furthermore, subjects claimed that by listening to the recommended music from the app for 5 minutes make them feel more relaxed than before.