Thesis (MPhil)--Stellenbosch University, 2012. / ENGLISH ABSTRACT: Tonality cognition incorporates a number of diverse and multidisciplinary aspects, including music cognition, acoustics, culture, computer-aided modelling, music theory and brain science. Current research shows growing emphasis on the use of computational models implemented on digital computers for music analysis, particularly with reference to the analysis of statistical properties, form and tonal properties. The applications of these analytical techniques are numerous, including the classification of genre and style, Music Information Retrieval (MIR), data mining and algorithmic composition. The research described in this document focuses on three aspects of tonality analysis, namely music cognition, computational modelling and music theory, particularly from the perspectives of statistical analysis and key-finding. Mathematical formulations are presented for a number of computational algorithms for analysing the statistical and tonal properties of music encoded in symbolic format. These include algorithms for determining the distributions of note durations, pitch intervals and pitch classes for statistical analysis and for template-based key-finding for tonal analysis. The implementation and validation of these computational algorithms on the Matlab software platform are subsequently discussed. The software application is used to determine whether a more optimal combination of pitch class weighing model and key profile template for the template-based key-finding algorithm can be derived, using the 24 preludes from Bach’s Well-tempered Clavier Book I, the Courante from Bach's Cello Suite in C major and the Gavotte from Bach's French Suite No. 5 in G major (BWV 816) as test material. Four pitch class weighing models, namely histogram weighing, flat weighing, linear durational weighing and durational accent weighing, are investigated. Two prominent key profile templates proposed in literature are considered, namely a key profile derived from tonality cognition experiments and a key profile based on classical music theory principles. The results show that the key-finding performances of all the combinations of the pitch class weighing models and existing key profile templates depend on the nature of the test material and that none of the combinations perform optimally for all test material. The software application is subsequently used to determine whether a more optimal key profile template can be derived using a pattern search parameter estimation algorithm. This investigation was conducted for diverse sets of search conditions, including unconstrained and constrained key profile coefficients, different pitch class weighing models, various key resolutions and different search algorithm parameters. Using the same sample material as for the key-finding evaluations, the investigation showed that a more optimal key profile, compared to existing profiles, can be derived. In comparing the average key-finding scores for all of the test material, the optimised profiles outperform the existing profiles substantially. The optimised key profiles introduce new pitch class hierarchies where the supertonic and the subdominant rate higher at the expense of the mediant in the major profile to improve the tracking of key modulations. / AFRIKAANSE OPSOMMING: Kognitiewe tonaliteit behels 'n aantal uiteenlopende en multidissiplinêre aspekte, insluitende musiek, akoestiek, kultuur, rekenaargesteunde modelering, musiekteorie en breinwetenskap. Huidige navorsing toon toenemende klem op die gebruik van berekenende modelering wat op digitale rekenaars geimplimenteer is vir musiekanalise, veral met verwysing na die analise van statistiese eienskappe, vorm en tonale eienskappe. Die aanwending van hierdie analitiese tegnieke is veelvoudig, insluitende die klassifikasie van genre of styl, onttrekking van musiekinformasie, dataversameling en algoritmiese komposisie. Die navorsing wat in hierdie dokument beskryf word fokus op drie aspekte van tonaliteit analise, naamlik musiekkognisie, berekenende modelering en musiekteorie, veral vanuit die perspektiewe van statistiese analise and toonsoortsoek. Wiskundige formulerings word aangebied vir 'n aantal berekeningalgoritmes vir die analise van die statistiese en tonale eienskappe van musiek wat in simboliese formaat ge-enkodeer is. Hierdie sluit algoritmes in vir die bepaling van die verspreidings van nootlengtes, toonintervalle en toonklasse vir statistiese analise en vir templaatgebaseerde toonsoortsoek vir tonale analise. Die implementering en validering van hierdie berekeningalgoritmes op die Matlab programmatuur platvorm word vervolgens bespreek. Die programmatuur toepassing word vervolgens gebruik om te bepaal of 'n meer optimale kombinasie van toonklas weegmodel en toonsoortprofiel templaat vir die templaat-gebaseerde toonsoortsoek algoritme afgelei kan word, deur gebruik te maak van Bach se Well-tempered Clavier Book I, die Courante van Bach se Cello Suite in C major en die Gavotte van Bach se French Suite No. 5 in G major (BWV 816) as toetsmateriaal. Vier toonklas weegmodelle, naamlik histogram weging, plat weging, lineêre duurtyd weging en duurtyd aksent weging, word ondersoek. Twee prominente toonsoortprofiel template uit die literatuur word oorweeg, naamlik 'n toonsoortprofiel wat van tonaliteit kognisie eksperimente afgelei is en 'n toonsoortprofiel gebaseer op klassieke musiekteoretiese beginsels. Die resultate wys dat die toonsoortsoek prestasies van al die kombinasies van die toonklas weegmodelle en bestaande toonsoortprofiel template afhang van die aard van die toetsmateriaal en dat geen van die kombinasies optimaal presteer vir alle toetsmateriaal nie. Die programmatuur toepassing word vervolgens aangewend om vas te stel of 'n meer optimale toonsoortprofiel afgelei kan word deur gebruik te maak van 'n patroonsoek parameterestimasie algoritme. Hierdie ondersoek is uitgevoer vir uiteenlopende stelle soektoestande, insluitende onbeperkte en beperkte toonsoortprofiel koëffisiënte, verskillende toonklas weegmodelle, 'n verskeidenheid toonsoort resolusies en verskillende soekalgoritme parameters. Deur gebruik te maak van dieselfde toetsmateriaal as vir die toonsoortsoek evaluerings, toon die ondersoek dat 'n meer optimale toonsoortprofiel, in vergelyking met bestaande profiele, afgegelei kan word. In 'n vergelyking van die gemiddelde toonsoortsoek prestasie vir al die toetsmateriaal, presteer die geoptimeerde profiele aansienlik beter as die bestaande profiele. The ge-optimeerde toonsoortprofiele lei tot nuwe toonklas hiërargiee waar die supertonikum en die subdominant hoër rangposissies beklee ten koste van die mediant in die majeur profiel, ten einde die navolg van toonsoort modulasies te verbeter.

This dissertation features methods of analyzing symbolic music, focused on n-gram-based approaches, as this representation resembles the most text and natural languages. The analysis of similarities between several text and music corpora is accompanied with implementation of text-based methods for problems of composer classification and symbolic music similarity definition. Both problems contain thorough evaluation of performance of the systems with comparisons to other approaches on existing testbeds. It is also described how one can use this symbolic representation in conjunction with genetic algorithms to tackle problems like melody generation. The proposed method is fully automated, and the process utilizes n-gram statistics from a sample corpus to achieve it. A method of visualization of complex symbolic music pieces is also presented. It consist of creating a self similarity matrix of a piece in question, revealing dependencies between voices, themes and sections, as well as music structure. A fully automatic technique of inferring music structure from these similarity matrices is also presented The proposed structure analysis system is compared against similar approaches that operate on audio data. The evaluation shows that the presented structure analysis system outperformed significantly all audio-based algorithms available for comparison in both precision and recall.

Similarity in music is a concept with significant impact on ethnomusicology studies, music recommendation systems, and music information retrieval systems such as Shazam and SoundHound. Various computer-based melody similarity measures have been proposed, but comparison and evaluation of similarity measures is inherently difficult due to the subjective and application-dependent nature of similarity in music. In this thesis, we address the diversity of the problem by defining a set of music transformations that provide the criteria for comparing and evaluating melody similarity measures. This approach provides a flexible and extensible method for characterizing selected facets of melody similarity, because the set of music transformations can be tailored to the user and to the application. We demonstrate this approach using three music transformations (transposition, tempo rescaling, and selected forms of ornamentation) to compare and evaluate several existing similarity measures, including String Edit Distance measures, Geometric measures, and N-Gram based measures. We also evaluate a newly implemented distance measure, the Beat and Direction Distance Measure, which is designed to have greater awareness of the beat hierarchy and better responsiveness to ornamentation. Training and test data is drawn from music incipits from the RISM A/II collection, and ground truth is taken from the MIREX 2005 Symbolic Melodic Similarity task. Our test results show that similarity measures that are responsive to music transformations generally have better agreement with human generated ground truth. / Thesis (Master, Computing) -- Queen's University, 2012-08-31 11:03:01.167

No description available.

Ce mémoire traite des modèles génératifs profonds appliqués à la génération automatique de musique symbolique. Nous nous attacherons tout particulièrement à concevoir des modèles génératifs interactifs, c'est-à-dire des modèles instaurant un dialogue entre un compositeur humain et la machine au cours du processus créatif. En effet, les récentes avancées en intelligence artificielle permettent maintenant de concevoir de puissants modèles génératifs capables de générer du contenu musical sans intervention humaine. Il me semble cependant que cette approche est stérile pour la production artistique dans le sens où l'intervention et l'appréciation humaines en sont des piliers essentiels. En revanche, la conception d'assistants puissants, flexibles et expressifs destinés aux créateurs de contenus musicaux me semble pleine de sens. Que ce soit dans un but pédagogique ou afin de stimuler la créativité artistique, le développement et le potentiel de ces nouveaux outils de composition assistée par ordinateur sont prometteurs. Dans ce manuscrit, je propose plusieurs nouvelles architectures remettant l'humain au centre de la création musicale. Les modèles proposés ont en commun la nécessité de permettre à un opérateur de contrôler les contenus générés. Afin de rendre cette interaction aisée, des interfaces utilisateurs ont été développées ; les possibilités de contrôle se manifestent sous des aspects variés et laissent entrevoir de nouveaux paradigmes compositionnels. Afin d'ancrer ces avancées dans une pratique musicale réelle, je conclue cette thèse sur la présentation de quelques réalisations concrètes (partitions, concerts) résultant de l'utilisation de ces nouveaux outils. / This thesis discusses the use of deep generative models for symbolic music generation. We will be focused on devising interactive generative models which are able to create new creative processes through a fruitful dialogue between a human composer and a computer. Recent advances in artificial intelligence led to the development of powerful generative models able to generate musical content without the need of human intervention. I believe that this practice cannot be thriving in the future since the human experience and human appreciation are at the crux of the artistic production. However, the need of both flexible and expressive tools which could enhance content creators' creativity is patent; the development and the potential of such novel A.I.-augmented computer music tools are promising. In this manuscript, I propose novel architectures that are able to put artists back in the loop. The proposed models share the common characteristic that they are devised so that a user can control the generated musical contents in a creative way. In order to create a user-friendly interaction with these interactive deep generative models, user interfaces were developed. I believe that new compositional paradigms will emerge from the possibilities offered by these enhanced controls. This thesis ends on the presentation of genuine musical projects like concerts featuring these new creative tools.