Automatic Music Mood Detection Using Transfer Learning and Multilayer Perceptron

Abstract

This paper proposes an automatic mood detection of music with a composition of transfer learning and multilayer. The five layered convolutional neural network pre-trained on Million Song dataset is used to extract the features from EmoMusic dataset. We obtain a set of features from the different five layers, which is fed into multilayer perceptron (MLP)-based regression. Through the regression network we estimate the mood of music on Thayer's two-dimensional emotion space, which consists of the axes corresponding to arousal and valence. Because the EmoMusic dataset does not provide enough number of data for training, we augment the data by time stretching to make it tripled. We perform the experiment with the augmented data as well as the original EmoMusic dataset. Box and whisker plot along with the mean of 10-fold cross-validation has been used for evaluating the proposed mood detection. In terms of the percentage of R 2 score for measure of accuracy, the proposed MLP shows state-of-the-art estimates for the augmented EmoMusic dataset.

Music Recommendations Based on User's Mood Using Convolutional Neural Networks

“…This approach was utilized before by other authors such as Delbouys et al [3], for predicting emotions using CNNs. Tables 5, 6 and 7 compare our results 3) on the testing data to three similar approaches, those of Bhattarai et al [1], aforementioned Delbouys et al [3] and Malik et al [10]. The last approach is the most similar to our hybrid network.…”

“…Max-Pooling layers downsizes the input using the maximum values from a stride. A stride is a portion from a matrix having a fixed size [1].…”

“…However, these approaches are based on labeling music by their emotions. Two state-of-the-art methods using CNNs, that use the same data representation as our work, consist in the works of Delbouys et al [3] and Bhattarai et al [1]. These methods implement a fully-connected network as the final layer, which will output the prediction.…”

“…These features are further utilized in Deep Neural Network architectures to predict the results. In most studies, deep learning approaches outperform traditional machine learning models [1].…”

Evaluation of the Emotion Model in Electronic Music Based on PSO-BP

“…There are some similarities between music emotion classification and analysis research and speech emotion analysis, and the biggest difference between the two is that the duration of music is longer, and the content composition is more complex. If emotion analysis needs to extract a large number of emotional features from music, it has many dimensions and has high difficulty in feature analysis [ 4 ]. The key influencing factors of music emotion analysis results are characteristic parameters and classification methods, which is also an important research direction for most scholars.…”