time slice transmitter by Marco Döttlinger

time slice transmitter is composed as an intra-active adaptive assemblage and investigates creative agency in the context of electroacoustic and algorithmic real-time composition and performance. Marco Döttlinger and Anna Lindenbaum (violin) explore process-based forms or relationships between human and algorithmic actors. This structure consists of three interlocking recurrent levels: a violinist, an autonomously acting computer music system (CMS) and a laptop performer. Our hypotheses are that musical intra-action with algorithms generates a new ontology of musical artworks: away from a text to be interpreted, towards a non-linear and radically open-ended workflow. This also calls into question the traditional concept of authorship, as the temporal interdependence represents new constellations of cooperation. In general, machine learning algorithms are pushing for a critical and creative examination of traditional concepts of music creation.

time slice transmitter TIMELINE

This exposition documents the intra-active adaptive research process. Each plot below with sound examples on the right represents the state of the system.

time slice transmitter is composed as a trio structure - in regularly recurring sessions we enter this structure and improvise. This open-ended working process is the prerequisite for engaging with the appeals of this structure, for acting or reacting musically. What is essential is the hypothesis that the composition exists in the design of the structure, which confronts us as a black box. We explore this space of possibilities in iterations and the timeline on this page shows the course of the project over the next year.

The Computer-Music-System (CMS) uses machine learning techniques and is corpus-based. The initial dataset is the audio recordings of the web project, consisting of about two hours of composed violin passages. This corpus was segmented into many thousands of samples and its audio descriptors (timbre, chroma and others) were used for an unsupervised machine learning model (UMAP). Dimension reduction techniques make it possible to display the entire corpus as a plot in two dimensions (see below, this plots represent the current state of the CMS).

Based on this, the CMS can now imitate or counterpoint the live performance; the accuracy depends on whether the violinist presents the CMS with familiar or unfamiliar material. Two further neural networks (regression and classification) were trained with two-dimensionally projected gestures from the violin data set. These models calculate predictions, i.e. possible continuations of the live performance.

Each session is recorded and the violin passages invented by the musician are added to the training datasets. The CMS thus gradually becomes richer in possible tonal but also temporal expressions, which in turn are informed by all the violinist's previous passages. This feedback loop is by no means objective or neutral, a notion that is often prevalent in connection with data acquisition. Our approach seeks to understand this very data capture as an inherently aesthetic and material music-making practice, a process-based act of musical communication. The chosen machine learning techniques (model types, choice of features, etc.) are to be understood as compositional decisions.

Public Performances with TST Version 1:

AUG 03 2024 - Noiz//Elektrorauschen, Waltherpark, Innsbruck (AT)
AUG 09 2024 - Kultursommer Wien, Währingerpark, Vienna (AT)
AUG 10 2024 - Exhibition "Land(wirt)schaft oder: it goes not on a cow hide“, Galerie Mauracherhof, Kitzbühel (AT)
NOV 08 2024 - Cafe am Heumarkt, Vienna (AT)

Audio-Examples coming soon!

Tip

Tip

time slice transmitter TIMELINE