A study on n-gram indexing of musical features

Abstract

Since only simple symbol-based manipulations are needed, n-gram indexing is used for natural languages where syntactic or semantic analyses are often difficult. Music, whose automatic analysis for patterns such as motifs and phrases are difficult, inaccurate or computationally expensive, is thus similar to natural languages. The use of n-gram in music retrieval systems is thus a natural choice.In this paper, we study a number of issues regarding ngram indexing of musical features using simulated queries. They are: whether combinatorial explosion is a problem in n-gram indexing of musical features, the relative discrimination power of six different musical features, the value of n needed for them, and the average amount of false positives returned when n-grams are used to index music.

Content-based audio classification and retrieval

“…Chou et al [16] implemented a music database system based on the chord-representation model and the PAT-tree index structure with "unstructured search" ability, where a chord was a combination of three or more notes which sounded together in harmony and a PAT-tree was a Patricia-like tree constructed over all possible substrings of the chord. Lap and Kao [17] studied a number of issues regarding n-gram indexing of musical features using simulated queries. Yang [18] brought out another prototype method of indexing raw-audio music files in a way that facilitated content-based similarity retrieval.…”