Exploiting Synchronized Lyrics And Vocal Features For Music Emotion Detection

Support Vector Machines are employed engaging playlists according to sentiment and with good results also for multilabel classification [30], emotions. While previous works were mostly based more recently also Convolutional Neural Networks were on audio for music discovery and playlists generation, used in this field [45]. Lyrics-based approaches, on the we take advantage of our synchronized lyrics dataset other hand, make use of Recurrent Neural Networks architectures to combine text representations and music features in (like LSTM [13]) for performing text classification a novel way; we therefore introduce the Synchronized [46, 47]. The idea of using lyrics combined with Lyrics Emotion Dataset. Unlike other approaches that voice only audio signals is done in [29], where emotion randomly exploited the audio samples and the whole recognition is performed by using textual and speech data, text, our data is split according to the temporal information instead of visual ones. Measuring and assigning emotions provided by the synchronization between lyrics to music is not a straightforward task: the sentiment/mood and audio. This work shows a comparison between associated with a song can be derived by a combination of text-based and audio-based deep learning classification many features, moreover, emotions expressed by a musical models using different techniques from Natural Language excerpt and by its corresponding lyrics do not always Processing and Music Information Retrieval domains.