Real-valued feature indexing for music databases

Abstract

The management of large collections of music data in a multimedia database has received much attention in the past few years. In the most of current works, the researchers extract the features, such as melodies, rhythms and chords, from the music data and develop indices that will help to retrieve the relevant music quickly. Several reports have pointed out that these features of music data can be transformed and represented in the forms of music feature strings or numeric values such that string indexing or numeric indexing is created, respectively, for music retrieval. For string indexing, there is only limited index structure (ex. suffix tree) suitable for music retrieval and it is lack of scalability. Moreover, for numeric indexing, there is only few research emphasized on this issue. The existing approaches all transform a specific length of music segments (features) into integers such that various numeric index structures can be applied (ex. R-tree, Btree). In this approach, however, the length of query (query by example) is required to match the specific length of transformation of music data otherwise it will harm the efficiency of query processing. To address these problems, in this paper, we will present a real value transformation function for without specific length of music segment and for more flexible of query length. Our experimental results also show that the new approach outperforms existing index schemes.

Approximate Searching for Music Data in Real-Valued Feature Indexing1

“…These 19 notes are low-pitched Mi~Si, mediant Do~Si, and high-pitched Do~Si which are represented by 'a'~'e', 'f'~'l', and 'm'~'s', respectively. The mediant Sol (marked as 'j') is the highest occurrence frequency note [20]. We also find that the appearance curve demonstrates a normal distribution of music notes.…”

“…Such processing will harm the efficiency of query processing. To address this problem, Lo and Tsai proposed real-valued indexing technique for music data in 2009 [20]. In this approach, a mapping function is provided for converting a music feature string into a real value.…”

“…In this approach, however, the length of a query (query by example) is required to match the specific length of transformation of music data otherwise it will harm the efficiency of query processing. A real-valued feature indexing for music databases was proposed by Lo and Tsai [20]. This approach is more flexible in query length and more Approximate Searching for Music Data in Real-Valued Feature Indexing1…”

Comparison Of Modified Dual Ternary Indexing And Multi-Key Hashing Algorithms For Music Information Retrieval

“…The fully indexed songs gave better results irrespective of the query length and the position of the query fragment. Another work has been proposed by Yu-lung Lo et al for indexing and retrieval to address the problem of variable query length of the music segment during retrieval [8]. The authors have used existing indexing techniques in terms of suffix trees to match the query phrase into a tree representation and have addressed the varying query length of the segment.…”