With the growth in digital representations of music, and of music stored in these representations, it is increasingly attractive to search collections of music. One mode of search is by similarity, but, for music, similarity search presents several difficulties: in particular, deciding what part of the music is likely to be perceived as the theme by a listener, and deciding whether two pieces of music with different sequences of notes represent the same theme. In this paper we propose a three-stage framework for matching pieces of music. We use the framework to compare a range of techniques for determining whether two pieces of music are similar, by experimentally testing their ability to retrieve different transcriptions of the same piece of music from a large collection of MIDI files. These experiments show that different comparison techniques differ widely in their effectiveness; and

Large volumes of music are available online, represented in performance formats such as MIDI and, increasingly, in abstract notation such as SMDL. Many types of user would find it valuable to search collections of music via queries representing music fragments, but such searching requires a reliable technique for identifying whether a provided fragment occurs within a piece of music. The problem of matching fragments to music is made difficult by the psychology of music perception, because literal matching may have little relation to perceived melodic similarity, and by the interactions between the multiple parts of typical pieces of music. In this paper we analyse the properties of music, music perception, and music database users, and use the analysis to propose alternative techniques for extracting monophonic melodies from polyphonic music; we believe that such melodies can subsequently be used for matching of queries to data. We report on experiments with music listeners, which rank our proposed techniques for extracting melodies.

In order to represent musical content, pitch and timing information is utilized in the majority of existing work in Symbolic Music Information Retrieval (MIR). Symbolic representations such as MIDI allow the easy calculation of such information and its manipulation. In contrast, most of the existing work in Audio MIR uses timbral and beat information, which can be calculated using automatic computer audition techniques. In this paper, Pitch Histograms are defined and proposed as a way to represent the pitch content of music signals both in symbolic and audio form. This representation is evaluated in the context of automatic musical genre classification. A multiple-pitch detection algorithm for polyphonic signals is used to calculate Pitch Histograms for audio signals. In order to evaluate the extent and significance of errors resulting from the automatic multiple-pitch detection, automatic musical genre classification results from symbolic and audio data are compared. The comparison indicates that Pitch Histograms provide valuable information for musical genre classification. The results obtained for both symbolic and audio cases indicate that although pitch errors degrade classification performance for the audio case, Pitch Histograms can be effectively used for classification in both cases. 1.

In a music retrieval system, a user presents a piece of music as a query and the system must identify from a corpus of performances other pieces with a similar melody. Several techniques have been proposed for matching such queries to stored music. In previous work, we found that local alignment, a technique derived from bioinformatics, was more effective than the n-gram methods derived from information retrieval; other researchers have reported success with n-grams, but have not compared against local alignment. In this paper we explore a broader range of n-gram techniques, and test them with both manual queries and queries automatically extracted from MIDI files. Our experiments show that n-gram matching techniques can be as effective as local alignment; one highly effective technique is to simply count the number of n-grams in common between the query and the stored piece of music. N-grams are particularly effective for short queries and manual queries, while local alignment is superior for automatic queries.

This paper proposes a novel framework for music content indexing and retrieval. The music structure information, i.e., timing, harmony and music region content, is represented by the layers of the music structure pyramid. We begin by extracting this layered structure information. We analyze the rhythm of the music and then segment the signal proportional to the inter-beat intervals. Thus, the timing information is incorporated in the segmentation process, which we call Beat Space Segmentation. To describe Harmony Events, we propose a two-layer hierarchical approach to model the music chords. We also model the progression of instrumental and vocal content as Acoustic Events. After information extraction, we propose a vector space modeling approach which uses these events as the indexing terms. In queryby-example music retrieval, a query is represented by a vector of the statistics of the n-gram events. We then propose two effective retrieval models, a hard-indexing scheme and a soft-indexing scheme. Experiments show that the vector space modeling is effective in representing the layered music information, achieving 82.5 % top-5 retrieval accuracy using 15-sec music clips as the queries. The soft-indexing outperforms hard-indexing in general.

Abstract. Content-based music information retrieval uses features extracted from music to answer queries. For melodic queries, the two main features are the pitch and duration of notes. The note pitch feature has been well researched whereas duration has not been fully explored. In this paper, we discuss how the note duration feature can be used to alter music retrieval effectiveness. Notes are represented by strings called standardisations. A standardisation is designed for approximate string matching and may not capture melodic information precisely. To represent pitches, we use a string of pitch differences. Our duration standardisation uses a string of five symbols representing the relative durations of adjacent notes. For both features, the Smith-Waterman alignment is used for matching. We demonstrate combining the similarity in both features using a vector model. Results of our experiments in retrieval effectiveness show that note duration similarity by itself is not useful for effective music retrieval. Combining pitch and duration similarity using the vector model does not improve retrieval effectiveness over the use of pitch on its own. 1

We have explored methods for music information retrieval for polyphonic music stored in the MIDI format. These methods use a query, expressed as a series of notes that are intended to represent a melody or theme, to identify similar pieces. Our work has shown that a three-phase architecture is appropriate for this task, in which the first phase is melody extraction, the second is standardisation, and the third is query-to-melody matching. We have investigated and systematically compared algorithms for each of these phases. To ensure that our results are robust, we have applied methodologies that are derived from text information retrieval: we developed test collections and compared different ways of acquiring test queries and relevance judgements. In this paper we review this program of work, compare to other approaches to music information retrieval, and identify outstanding issues.

Most research into music information retrieval thus far has only examined music from the western tradition. However, music of other origins often conforms to different tuning systems. Therefore there are problems both in representing this music as well as finding matches to queries from these diverse tuning systems. We discuss the issues associated with microtonal music retrieval and present some preliminary results from an experiment in applying scoring matrices to microtonal matching. 1.

Time series representations are common in MIR applications such as query-by-humming, where a sung query might be represented by a series of `notes' for database retrieval. While such a transcription into (pitch, duration) pairs is convenient and musically intuitive, there is no evidence that it is an optimal representation. The present work explores three time series representations for sung queries: a sequence of notes, a `smooth' pitch contour, and a novel sequence of pitch histograms. Dynamic alignment procedures are described for the three representations. Multiple continuity constraints are explored and a modified dynamic alignment procedure is described for the histogram representation. We measure the performance of the three representations using a collection of naturally sung queries applied to a target database of varying size. The results show that the note representation lends itself to rapid retrieval whereas the contour representation lends itself to robust performance. The histogram representation yields performance nearly as robust as the contour representation, but with computational complexity similar to the note representation.

In this paper, we identify parameters crucial to the performance of a Query By Humming (QBH) system, and present an analytical approach to determining optimal values of such parameters. Existing systems use heuristically chosen parameters- our analytical results are in accordance with such values. We present results of experimentation with simulated data, as well as an actual melody database of a QBH system. 1.