MUSE prototype for Music Sentiment Expression
Over the past few years, text-based sentiment analysis tools have evolved into mature services and APIs. For instance, tools like LIWC (Linguistic Inquiry – Word Count) and IBM’s ToneAnalyzer can extract sentiments from texts to report and predict expected user feedback However, very few comparable breakthroughs have been made when it comes to analyzing multimedia documents (e.g., images, sounds, and videos).
Musical Sentiment Analysis (MSA) attempts to bridge this gap between text and music. Given an input musical piece, an MSA tool should accurately estimate end users’ emotional response when listening to the given piece. The potential applications of such a sentiment analysis system are broad and could have a serious impact in the field. For one, it could help music producers gauge their compositions to check whether they will produce the target sentiments they were attempting to portray. Beyond that, it could usher in a new sentiment-based music search functionality, in which musical pieces are retrieved based on their expected sentiment vectors. Most importantly, it could herald the start of the development of a universal retrieval system, where any multimedia document of any type (including images, videos, music, etc.) could be retrieved based on its perceived sentiment vector, irrespective of the media-specific features (e.g., visual, musical, spectral) that are part of its nature, and which are only dealt with at the sentiment-analysis stage.
In this paper, we concisely describes and evaluates our musical sentiment analysis prototype system titled MUSE (MUsic Sentiment Extraction), taking as input a MIDI music file and producing as output a sentiment vector describing the 6 primary emotions (i.e., anger, fear, joy, love, sadness, and surprise) expressed by the music file.
Fig. 1. Sentiment Engine Architecture