Combination of audio and lyrics features for genre classification in digital audio collections

Abstract

In many areas multimedia technology has made its way into mainstream. In the case of digital audio this is manifested in numerous online music stores having turned into profitable businesses. The widespread user adaption of digital audio both on home computers and mobile players show the size of this market. Thus, ways to automatically process and handle the growing size of private and commercial collections become increasingly important; along goes a need to make music interpretable by computers. The most obvious representation of audio files is their sound -there are, however, more ways of describing a song, for instance its lyrics, which describe songs in terms of content words. Lyrics of music may be orthogonal to its sound, and differ greatly from other texts regarding their (rhyme) structure. Consequently, the exploitation of these properties has potential for typical music information retrieval tasks such as musical genre classification; so far, there is a lack of means to efficiently combine these modalities. In this paper, we present findings from investigating advanced lyrics features such as the frequency of certain rhyme patterns, several parts-of-speech features, and statistic features such as words per minute (WPM). We further analyse in how far a combination of these features with existing acoustic feature sets can be exploited for genre classification and provide experiments on two test collections.

Improving automatic music genre classification with hybrid content-based feature vectors

“…By combining SSD and Symbolic features they achieved an accuracy of 74.5%, an improvement of 3.1% in the classification accuracy. On a different database, but compatible in size with the LMD, [16] has achieved an accuracy of 66.32% with SSD features alone and an accuracy of 68.72%, an improvement of 2.40%, when combining SSD with features extracted from the song Lyrics. If we compare the best results obtained in the ISMIR genre database, we will see that by using only SSD we have an accuracy of 76.12%.…”

“…The additional sources of information that has been used to augment the content-based approaches are: Cultural information by using web-mining techniques [27]; Boolean meta-data tags representing the music context [1]; Lyrics [16]; Symbolic representation by using a transcription system [15]; and the combination of cultural information and symbolic representation features [18]. In all these approaches the use of additional sources of information has improved the classification accuracy when compared to using only the audio content-based features.…”

“…This is due to the fact that the music genre in one of the most frequent items used in search queries, as pointed by several authors [5,13,21]. Over the past decade this task has received growing attention every year which lead to the development of different types of feature descriptors [9,14,26] and classifications approaches [1,15,16,18,24]. Current research on automatic music genre classification suggests that the use of content-based approaches have reached a glass ceiling [15].…”

“…Current research on automatic music genre classification suggests that the use of content-based approaches have reached a glass ceiling [15]. In order to overcome this situation, several authors have proposed the use of different types of information related to the music pieces in combination with the content-based features [1,15,16,18,27].…”

Mood classifiaction of lyrics using SentiWordNet

“…There are many novel sets of style features for automatic lyrics processing. Rudolf Mayer [11] has presented features to capture rhyme, parts-of-speech, and text statistics characteristics for song lyrics. He has combined these new feature sets with the standard bag-of-words features and wellknown feature sets for acoustic analysis of digital audio tracks.…”