Love Me, Love Me, Say (and Write!) that You Love Me: Enriching the WASABI Song Corpus with Lyrics Annotations

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 2138–2147 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 2138 Love Me, Love Me, Say (and Write!) that You Love Me: Enriching the WASABI Song Corpus with Lyrics Annotations Michael Fell, Elena Cabrio, Elmahdi Korfed, Michel Buffa and Fabien Gandon Universit´e Cˆote d’Azur, CNRS, Inria, I3S, France {michael.fell, elena.cabrio, elmahdi.korfed, michel.buffa}@unice.fr, fabien.gandon@inria.fr Abstract We present the WASABI Song Corpus, a large corpus of songs enriched with metadata extracted from music databases on the Web, and resulting from the processing of song lyrics and from audio analysis. More specifically, given that lyrics encode an important part of the semantics of a song, we focus here on the description of the methods we proposed to extract relevant information from the lyrics, such as their structure segmentation, their topics, the explicitness of the lyrics content, the salient passages of a song and the emotions conveyed. The creation of the resource is still ongoing: so far, the corpus contains 1.73M songs with lyrics (1.41M unique lyrics) annotated at different levels with the output of the above mentioned methods. Such corpus labels and the provided methods can be exploited by music search engines and music professionals (e.g. journalists, radio presenters) to better handle large collections of lyrics, allowing an intelligent browsing, categorization and recommendation of songs. We provide the files of the current version of the WASABI Song Corpus, the models we have built on it as well as updates here: https://github.com/micbuffa/WasabiDataset. Keywords: Corpus (Creation, Annotation, etc.), Information Extraction, Information Retrieval, Music and Song Lyrics 1. Introduction Let’s imagine the following scenario: following David Bowie’s death, a journalist plans to prepare a radio show about the artist’s musical career to acknowledge his quali- ties. To discuss the topic from different angles, she needs to have at her disposal the artist biographical information to know the history of his career, the song lyrics to know what he was singing about, his musical style, the emotions his songs were conveying, live recordings and interviews. Similarly, streaming professionals such as Deezer, Spotify, Pandora or Apple Music aim at enriching music listening with artists’ information, to offer suggestions for listening to other songs/albums from the same or similar artists, or automatically determining the emotion felt when listening to a track to propose coherent playlists to the user. To sup- port such scenarios, the need for rich and accurate musi- cal knowledge bases and tools to explore and exploit this knowledge becomes evident. In this paper, we present the WASABI Song Corpus, a large corpus of songs (2.10M songs, 1.73M with lyrics) enriched with metadata extracted from music databases on the Web, and resulting from the processing of song lyrics and from audio analysis. The corpus contains songs in 36 different languages, even if the vast majority are in English. As for the songs genres, the most common ones are Rock, Pop, Country and Hip Hop. More specifically, while an overview of …

Content truncated to 500 words