O'Reilly logo

Music Emotion Recognition by Homer H. Chen, Yi-Hsuan Yang

Stay ahead with the world's most comprehensive technology and business learning platform.

With Safari, you learn the way you learn best. Get unlimited access to videos, live online training, learning paths, books, tutorials, and more.

Start Free Trial

No credit card required

Saunder January 24, 2011 10:39 book
12
Genre Classification and
Its Application to MER
Continuing the previous two chapters, this chapter describes the use of genre meta-
data to bridge the semantic gap between audio signal and emotion perception. While
lyrics are complementary to music signal and chords describe the harmonic progres-
sion and the tonal structure of a song, genre metadata have been utilized to describe
the intrinsic form of music. Genre and emotion provide complementary descriptions
of music content and often correlate with each other. The two-layer emotion clas-
sification scheme introduced in this chapter exploits such correlation and improves
the accuracy of emotion classification.
12.1 Motivation
Genre, by which a song is classified into classical, jazz, rock, hip-hop, country, etc.,
has been used to describe the intrinsic form of music for a long time [28]. A user
study shows that genre is the most popular cue for searching music, in addition to
other metadata such as artist name, song title, and lyrics [193]. Most music websites,
such as Last.fm [6] and All Music Guide (AMG) [1], also provide the genre metadata
of music recordings. A good deal of work has been done in automatic music genre
classification; see [276] for a comprehensive review.
Emotion, as one of the preeminent functions of music [146], is also an important
means for music classification. An emotion-based music retrieval system provides
users the functionality for retrieving music according to emotion. Compared with
genre classification, emotion classification is considered more challenging because
of the subtlety of emotion and the difficulty of collecting (subjective) annotation of
emotion.
197
Saunder January 24, 2011 10:39 book
198 Music Emotion Recognition
Genre and emotion provide complementary descriptions of music content and
often correlate with each other. For example, a rock song is often aggressive, whereas
a rhythm and blues (R&B) song is more likely to be sentimental. Despite the salient
correlation between genre and emotion, genre classification and emotion classifica-
tion are often studied separately without considering the interrelation.
This chapter describes a two-layer music emotion classification scheme that ex-
ploits the correlation between genre and emotion [358]. The genre metadata are used
to aid emotion classification because genre metadata are easier to collect and because
genre classification is relatively easier. Specifically, the genre of a song is predicted
in the first layer, and then the genre-specific emotion classification model is applied
in the second layer to predict the emotion of the song, as shown in Figures 12.1(b)
and (c). Experimental results are provided to show the superiority of the two-layer
music emotion classification scheme over the traditional single-layer scheme.
12.2 Two-Layer Music Emotion Classification
The main idea of the two-layer emotion classification scheme is to group songs by
genre and train an emotion classifier specifically for songs of each genre. The use of
such genre-specific classifiers is motivated by the following two observations. First,
since emotion and genre are correlated, we may set up different emotion priors for
each genre-specific classifier. For example, a rap song is less likely to be relaxing than
a jazz song. Second, as a happy song of rock music and a happy song of jazz music
may sound substantially different, emotion classification may become easier if each
genre-specific classifier only needs to focus on a single genre of music.
As shown in Figure 12.1(a), a typical single-layer emotion classification system
is composed of two parts: feature extraction and emotion classification. An emotion
Feature
extraction
Te st music Test music Test music
Emotion
classification
Genre
classification
Emotions
Genre
classification
Emotion
classification
Probability of
being each genre
Emotions
Weighted
combination
Genre-specific
classifier
choosing
Emotion
classification
Emotions
Feature
extraction
Feature
extraction
(a) (b) (c)
Figure 12.1 Schematic diagrams of (a) the traditional single-layer emotion clas-
sification scheme, (b) the proposed two-layer scheme hardGenre, and (c) the soft
version of the two-layer scheme softGenre.

With Safari, you learn the way you learn best. Get unlimited access to videos, live online training, learning paths, books, interactive tutorials, and more.

Start Free Trial

No credit card required