The Cover Song Variation Dataset

As digital music collections grow larger, music similarity becomes one of the most prominent concepts in the field of Music Information Retrieval (MIR). Modelling similarity between music pieces allows efficient retrieval and organizing of such collections. Studies have shown that the concept of variation is closely related to similarity, since listeners tend to cluster together musical patterns that are repeated, transformed but still recognizable. Subsequently, musical pieces or segments that contain such patterns are considered similar. Such structural variations are notably present in oral-transmission processes. Folk songs are a standing example of such a process, capturing a huge amount of varying patterns moulded through time. Variations in cover songs in western popular music are also very interesting examples, since a) they can be considered products of a “modern” oral-transmission procedure and b) covers themselves are typically well documented with rich metadata