(2012) Utrecht College Repository
Supervisor(s): Veltkamp, Remco Wiering, Frans
Using the emergence of huge scale digitalisation of music, content-based techniques to maintain, structure, and supply use of digital music repositories have grown to be more and more important. This doctorate dissertation covers an array of techniques that aim to assist in the organisation of music information. From both an operating also . find out more as cognitive perspective, it’s logical to structure musical content by defining similarity relations between documents. Consequently, the idea of music similarity has turned into a fundamental concept inside the area music information retrieval (MIR) research. Within this dissertation we practice a particular kind of music similarity: the similarity of musical harmony. Because both musically trained and untrained listeners have extensive understanding about music, it is quite unlikely that information required for seem similarity judgement are available in the musical information source alone. Therefore, so that you can place chord sequences poor Western tonal harmony, we investigate two approaches towards automatic harmony analysis. Even though the first generative grammar-based solution yields great results on the small dataset, it uncovered some practical challenges that avoided so that it is extended to process bigger datasets. Hence, the 2nd harmonic analysis solution exploits condition-of-the-art functional programming techniques, like type-level computation and error-correcting parsers, to satisfy these challenges. This model, named HarmTrace, is fast, flexible, and returns analyses which are in compliance with harmony theory.
We evaluate these harmonic annotations, which explain the function of the chord in the tonal context, both qualitatively in addition to quantitatively, and show how they may help with harmonic similarity estimation and automatic chord transcription. We investigate three novel methods to harmonic similarity: a geometrical, a nearby alignment, along with a common embeddable subtree based approach. The geometric approach, named TPSD, utilizes a music theoretically motivated step operates to measure the similarity of two chord sequences the most popular embeddable subtree approach estimates harmonic similarity by matching hierarchical harmonic analysis annotations and also the local alignment solution uses context-aware substitution operates to align sequences of guitar chords. For all these harmonic similarity solutions, the adjustable parameters are discussed and evaluated. For that evaluation a sizable new chord sequence corpus is put together composed of 5028 different chord sequences, most of which describe exactly the same song. The outcomes reveal that an alignment approach that utilizes the HarmTrace harmony model performs very best in retrieving these similar chord sequences. All suggested similarity measures depend around the accessibility to sequences of symbolic chord labels. To increase the applying domain, we demonstrate how automatic chord transcription from musical audio could be improved by exploiting our type of tonal harmony show less
Open Access version via Utrecht College Repository
Writer: Utrecht College