Title: Music Trajectory; Deep Learning with Audio

Abstract

Generative adversarial networks (GANs) have been extensively investigated to transform perceptual and interpretative features of audio data into a latent space. Representing the characteristics of the training data-set, high-dimensional nature of a GAN latent space opens up a wide range of possibilities for exploring its sound world and interfacing to musically interesting compositions and performances. In this seminar talk, I will present our own experience of applying GAN architecture to the development of the AI-terity musical instrument. Through an iterative process we created an instrument that allows musicians to navigate the GAN latent space in a more controlled manner by interacting with its parameter space, deforming and manipulating its non-rigid interface. We have been investigating the possibility to generate musically complex sounds and develop more expressive and controllable performances through this approach. In this process, we write compositions for the AI-terity instrument that comprise GAN's computational features, evolving from sequences of sounds, rhythms and glitches, to structural forms with many timbral variations. Musical applications of deep learning models are extremely diverse and depend on many factors, such as a desired perceptual effect, musical genre, the architecture of the neural network, the dataset used for training, the choice of input features. In this part of the talk I will present the studio sessions that we recorded live in March 2022 with the AI-terity musical instrument. These studio sessions present a way of composing and performing in which musicians have been directed to explore, through AI-terity instrument, an unknown music trajectory, where a large set of new sound events can be discovered and performed. I will talk about musical results we obtained in these studio session, with the focus on the musicians’ experiences and impressions.

Bio

Koray Tahiroğlu is a musician, Academy Research Fellow and lecturer in the Department of Art and Media, Aalto University School of ARTS. He is the founder and head of SOPI (Sound and Physical Interaction) research group, coordinating research projects with interests including embodied approaches to sonic interaction, new interfaces for musical expression, deep learning and artificial intelligence (AI) technologies with audio. Since 2004, he has been also teaching workshops and courses introducing digital tools, computational environments, artistic strategies for music compositions and performances. Tahiroğlu has performed music in collaboration as well as in solo performances in Europe, North America and Australia. His work has been presented in venues, such as Ars Electronica, Sónar, AI x Music Festival, STEIM, TodaysArt and Audio Art Festival. In 2018, he was awarded a 5-year Academy of Finland Research Fellowship.