nach oben

2020 | Buch

Kapitel lesen Erstes Kapitel lesen

Artificial Intelligence in Music, Sound, Art and Design

9th International Conference, EvoMUSART 2020, Held as Part of EvoStar 2020, Seville, Spain, April 15–17, 2020, Proceedings

herausgegeben von: Juan Romero, Anikó Ekárt, Tiago Martins, João Correia

Verlag: Springer International Publishing

Buchreihe : Lecture Notes in Computer Science

Enthalten in: Springer Professional "Wirtschaft+Technik" , Springer Professional "Technik" , Springer Professional "Wirtschaft"

Einloggen, um Zugang zu erhalten

Über dieses Buch

This book constitutes the refereed proceedings of the 9th European Conference on Artificial Intelligence in Music, Sound, Art and Design, EvoMUSART 2020, held as part of Evo*2020, in Seville, Spain, in April 2020, co-located with the Evo*2020 events EuroGP, EvoCOP and EvoApplications.
The 15 revised full papers presented were carefully reviewed and selected from 31 submissions. The papers cover a wide spectrum of topics and application areas, including generative approaches to music and visual art, deep learning, and architecture.

Inhaltsverzeichnis

Frontmatter

A Deep Learning Neural Network for Classifying Good and Bad Photos

Abstract

Current state-of-the-art solutions that automate the assessment of photo aesthetic quality use deep learning neural networks. Most of these networks are either binary classifiers or regression models that predict the aesthetic quality of photos. In this paper, we developed a deep learning neural network that predicts the opinion score rating distribution of a photo’s aesthetic quality. Our work focused on finding the best pre-processing method for improving the correlation between ground truth and predicted aesthetic rating distribution of photos in the AVA dataset. We investigated three ways of image resizing and two ways of extracting regions based on salience. We found that the best pre-processing method depended on the photos chosen for the training set.

Stephen Lou Banal, Vic Ciesielski

Adapting and Enhancing Evolutionary Art for Casual Creation

Abstract

Casual creators are creativity support tools designed for non-experts to have fun with while they create, rather than for serious creative production. We discuss here how we adapted and enhanced an evolutionary art approach for casual creation. Employing a fun-first methodology for the app design, we improved image production speed and the quality of randomly generated images. We further employed machine vision techniques for image categorisation and clustering, and designed a user interface for fast, fun image generation, adhering to numerous principles arising from the study of casual creators. We describe the implementation and experimentation performed during the first stage of development, and evaluate the app in terms of efficiency, image quality, feedback quality and the potential for users to have fun. We conclude with a description of how the app, which is destined for public release, will also be used as a research platform and as part of an art installation.

Simon Colton, Jon McCormack, Sebastian Berns, Elena Petrovskaya, Michael Cook

Comparing Fuzzy Rule Based Approaches for Music Genre Classification

Abstract

Most of the studies on music genre classification are focused on classification quality only. However, listeners and musicologists would favor comprehensible models, which describe semantic properties of genres like instrument or chord statistics, instead of complex black-box transforms of signal features either manually engineered or learned by neural networks. Fuzzy rules – until now not a widely applied method in music classification – offer the advantage of understandability for end users, in particular in combination with carefully designed semantic features. In this work, we tune and compare three approaches which operate on fuzzy rules: a complete search of primitive rules, an evolutionary approach, and fuzzy pattern trees. Additionally, we include random forest classifier as a baseline. The experiments were conducted on an artist-filtered subset of the 1517-Artists database, for which 245 semantic properties describing instruments, moods, singing style, melody, harmony, influence on listener, and effects were extracted to train the classification models.

Frederik Heerde, Igor Vatolkin, Günter Rudolph

Quantum Zentanglement: Combining Picbreeder and Wave Function Collapse to Create Zentangles®

Abstract

This paper demonstrates a computational approach to generating art reminiscent of Zentangles by combining Picbreeder with Wave Function Collapse (WFC). Picbreeder interactively evolves images based on user preferences, and selected image tiles are sent to WFC. WFC generates patterns by filling a grid with various rotations of the tile images, placed according to simple constraints. Then other images from Picbreeder act as templates for combining patterns into a final Zentangle image. Although traditional Zentangles are black and white, the system also produces color Zentangles. Automatic evolution experiments using fitness functions instead of user selection were also conducted. Although certain fitness functions occasionally produce degenerate images, many automatically generated Zentangles are aesthetically pleasing and consist of naturalistic patterns. Interactively generated Zentangles are pleasing because they are tailored to the preferences of the user creating them.

Anna Krolikowski, Sarah Friday, Alice Quintanilla, Jacob Schrum

Emerging Technology System Evolution

Abstract

This paper introduces a strategy for exploring possibility spaces for creative design of emerging technology systems. These systems are represented as directed acyclic graphs with nodes representing different technologies and data or media types. The naming of system and data components is intended to be populated by a subjective, personal ontology. The source material can be simply modified by a user who wishes to explore a space of possibilities, constrained by specific resources, to generate simple visual descriptions of systems that can be quickly browsed. These dataflow visualizations are a familiar representation to those who use visual programming interfaces for creative media processing. An interactive genetic programming based approach is used with small populations of individual designs. The system is being developed using web browser technology. A minimum functionality proof of concept system has been implemented. Next steps and current challenges are discussed.

Matthew Lewis

Fusion of Hilbert-Huang Transform and Deep Convolutional Neural Network for Predominant Musical Instruments Recognition

Abstract

As a subset of music information retrieval (MIR), predominant musical instruments recognition (PMIR) has attracted substantial interest in recent years due to its uniqueness and high commercial value in key areas of music analysis such as music retrieval and automatic music transcription. With the attention paid to deep learning and artificial intelligence, they have been more and more widely applied in the field of MIR, thus making breakthroughs in some sub-fields that have been stuck in the bottleneck. In this paper, the Hilbert-Huang Transform (HHT) is employed to map one-dimensional audio data into two-dimensional matrix format, followed by a deep convolutional neural network developed to learn affluent and effective features for PMIR. In total 6705 audio pieces including 11 musical instruments are used to validate the efficacy of our proposed approach. The results are compared to four benchmarking methods and show significant improvements in terms of precision, recall and F1 measures.

Xiaoquan Li, Kaiqi Wang, John Soraghan, Jinchang Ren

Genetic Reverb: Synthesizing Artificial Reverberant Fields via Genetic Algorithms

Abstract

We present Genetic Reverb, a user-friendly vst 2 audio effect plugin that performs convolution with Room Impulse Responses (rirs) generated via a Genetic Algorithm (ga). The parameters of the plugin include some of the standard room acoustics parameters mapped to perceptual correlates (decay time, intimacy, clarity, warmth, among others). These parameters provide the user with some control over the resulting rirs as they determine the fitness values of potential rirs. In the ga, these rirs are initially generated via a Gaussian noise method, and then evolved via truncation selection, multi-point crossover, zero-value mutation, and Gaussian mutation. These operations repeat until a certain number of generations has passed or the fitness value reaches a threshold. Either way, the best-fit rir is returned. The user can also generate two different rirs simultaneously, and assign each of them to the left and right stereo channels for a binaural reverberation effect. With Genetic Reverb, the user can generate and store new rirs that represent virtual rooms, some of which may even be impossible to replicate in the physical world. An original musical composition using the Genetic Reverb plugin is presented to demonstrate its applications. (The source code and link to the demo track is available at https://github.com/edward-ly/GeneticReverb).

Edward Ly, Julián Villegas

Portraits of No One: An Interactive Installation

Abstract

Recent developments on artificial intelligence expedited the computational fabrication of visual information, especially photography, with realism and easiness never seen before. In this paper, we present an interactive installation that explores the generation of facial portraits in the borderline between the real and artificial. The presented installation synthesises new human faces by recombining the facial features of its audience and displays them on the walls of the room The array of faces displayed in the installation space is contaminated with real faces to make people question about the veracity of the portraits they are observing. The photo-realism of the generated faces makes it difficult to distinguish the real portraits from the artificial ones.

Tiago Martins, João Correia, Sérgio Rebelo, João Bicker, Penousal Machado

Understanding Aesthetic Evaluation Using Deep Learning

Abstract

A bottleneck in any evolutionary art system is aesthetic evaluation. Many different methods have been proposed to automate the evaluation of aesthetics, including measures of symmetry, coherence, complexity, contrast and grouping. The interactive genetic algorithm (IGA) relies on human-in-the-loop, subjective evaluation of aesthetics, but limits possibilities for large search due to user fatigue and small population sizes. In this paper we look at how recent advances in deep learning can assist in automating personal aesthetic judgement. Using a leading artist’s computer art dataset, we use dimensionality reduction methods to visualise both genotype and phenotype space in order to support the exploration of new territory in any generative system. Convolutional Neural Networks trained on the user’s prior aesthetic evaluations are used to suggest new possibilities similar or between known high quality genotype-phenotype mappings.

Jon McCormack, Andy Lomas

An Aesthetic-Based Fitness Measure and a Framework for Guidance of Evolutionary Design in Architecture

Abstract

The authors present an interactive design framework for grammar evolution. A novel aesthetic-based fitness measure was introduced as guidance procedure. One feature of this guidance procedure is the initial input of a reference image. This image provides direction to the evolutionary design process by application of a similarity measure. A case study shows the interactive exploration of a set of 3D-shapes using the presented framework for guidance of evolutionary design in architecture. The aesthetic-based fitness measure combing quantitative and qualitative criteria was applied in this case study to evaluate form and function of architectural design solutions.

Manuel Muehlbauer, Jane Burry, Andy Song

Objective Evaluation of Tonal Fitness for Chord Progressions Using the Tonal Interval Space

Abstract

Chord progressions are core elements of Western tonal harmony regulated by multiple theoretical and perceptual principles. Ideally, objective measures to evaluate chord progressions should reflect their tonal fitness. In this work, we propose an objective measure of the fitness of a chord progression within the Western tonal context computed in the Tonal Interval Space, where distances capture tonal music principles. The measure considers four parameters, namely tonal pitch distance, consonance, hierarchical tension and voice leading between the chords in the progression. We performed a listening test to perceptually assess the proposed tonal fitness measure across different chord progressions, and compared the results with existing related models. The perceptual rating results show that our objective measure improves the estimation of a chord progression’s tonal fitness in comparison with existing models.

María Navarro-Cáceres, Marcelo Caetano, Gilberto Bernardes

Coevolving Artistic Images Using OMNIREP

Abstract

We have recently developed OMNIREP, a coevolutionary algorithm to discover both a representation and an interpreter that solve a particular problem of interest. Herein, we demonstrate that the OMNIREP framework can be successfully applied within the field of evolutionary art. Specifically, we coevolve representations that encode image position, alongside interpreters that transform these positions into one of three pre-defined shapes (chunks, polygons, or circles) of varying size, shape, and color. We showcase a sampling of the unique image variations produced by this approach.

Moshe Sipper, Jason H. Moore, Ryan J. Urbanowicz

Sound Cells in Genetic Improvisation: An Evolutionary Model for Improvised Music

Abstract

Musical improvisation and biological evolution are similarly based on the principles of unpredictability and adaptivity. Within this framework, this research project examines whether and how structures of evolutionary developmental logic can be detected and described in free improvisation. The underlying concept of improvisation is participative in nature and, in this light, contains similar generative strategies as there are in evolutionary processes. Further implications of the theory of evolution for cultural development in the concept of memetics and the form of genetic algorithms build an interdisciplinary network of different theories and methodologies, from which the proposed model of genetic improvisation emerges.

Sebastian Trump

Controlling Self-organization in Generative Creative Systems

Abstract

We present a new tool which simulates the development of Artificial Chemistries (AChems) to produce real-time imagery for artistic/entertainment purposes. There have been many such usages of complex systems (CSs) for artistic purposes, but deciding which parameters to use for such unpredictable systems can lead to a feeling of lack of control. For our purposes, we struggled to gain enough control over the AChem real-time image generation tool to accompany music in a video-jockeying application. To overcome this difficulty, we developed a general-purpose clustering approach that attempts to produce sets of parameter configurations which lead to maximally distinct visualisations, thus ensuring users feel that they have influence over the AChem when controlled with a suitable GUI. We present this approach and its application to controlling the development of AChems, along with the results from experiments with different clustering approaches, aided by both machine vision analysis and human curation. We conclude by advocating an overfitting approach supplemented by a final check by a designer, and discuss potential applications of this in artistic and entertainment settings.

Jonathan Young, Simon Colton

Emulation Games

See and Be Seen, an Subjective Approach to Analog Computational Neuroscience

Abstract

Emulation consists in imitating a thing, trying to equal or even improve it. Playing means doing something for fun and entertainment. Under these premises, this paper proposes to describe an emulation game that addresses an approach to the area of Computational Neuroscience. It is in this context where we are designing, performing and valuing certain technical devices that could reveal, or make effective, computational events associated with the emergence of consciousness. The conceptual framework, in which is situated our game, can be defined as a possible epistemogony, a space where the physiology of knowledge and the making are activated, and from which we intend to highlight certain conditions that make possible both the so-called scientific research, and art making.

Augusto Zubiaga, Lourdes Cilleruelo

Backmatter

Titel: Artificial Intelligence in Music, Sound, Art and Design
herausgegeben von: Juan Romero
Anikó Ekárt
Tiago Martins
João Correia
Verlag: Springer International Publishing
Electronic ISBN: 978-3-030-43859-3
Print ISBN: 978-3-030-43858-6
DOI: https://doi.org/10.1007/978-3-030-43859-3