I occasionally put on the hat of an artist-researcher, working with adaptive compositional techniques that seek to foster dialog between sounds, the individual and the dynamics of the individual with the collective. I have had the wonderful opportunity to collaborate on projects with immersive artist Jan Schacher, ethnographer Hanna-Maija Pääkkönen and digital artist Antti Ilvessuo.

Trained as an instrumentalist and composer, the focus of my work has shifted from using traditional techniques, to aleatoric and computation models that transform composition into a sculptural process that can adapt and morph dynamically in response to action, perception and interaction from the listener. New territories are only truly discovered by going there: My work often takes a piece of music from the traditional repertoire as its point of departure. It is a means of rediscovery -- recontextualizing the familiar and creating a fresh response to it. Its helpful to think of this process in terms of an analogy: If you make the same drive through a beautiful landscape everyday, you don't notice the landscape after a while. The aim of my work is to take a detour through this familiar landscape to rediscover it in new ways.

An atomic example of the process of using computational models to generate motivic material is presented below -

hothouse plant - counterpoint

hothouse plant is an example of using a Markov chain (à la Xenakis) to generate counterpoint: two or more musical lines which are harmonically interdependent yet independent in melodic contour.
The first section (I) is a three note motif that will be used as the source material (performed by a small viola ensemble).
A Markov chain is used to generate a repertoire of lines as counterpoint to (I). The lines that have the closest harmonic relationship to the three note motif in (I) retained. The result of this process produces the second section (II) with four independent lines (performed by a small string section). Finally, the third section (III) is a variation of (II), created by mutating material from the second section by adding some Gaussian noise to the selection process in (II).

The music generated by the aleatoric processes and models is merely a starting point.
It is followed by revisions, re-interpretations and several mutations to produce the pieces that follow.

selected works

Plath: A Catalogue of Sketches 2023-24

From an upcoming ballet production that weaves together the strands of Sylvia Plath's life, showcasing the conflicts, passions, and inner turmoils that form the central figure of her work. The work is a reflection on Plath's evocative imagery, confessional style, and exploration of themes such as identity, social roles and the human condition.

tuileries on a summer night
An excerpt from Plath: A Catalogue of Sketches. The themes and major motifs were written as a response to Plath's Journals. These were then developed into larger fragments of music using a set of Markov chains that were trained on Shostakovich's late quartets.
Pillars of Power 2023

A study of the manifestations of hierarchies in a society with dwindling articulation and negotiation between the individual and the collective. The performance, primarily spoken word, attempts to explore ideas around oppressive power dynamics, the privilege to participate, and the legacy of the voices that dominate the conversations and critique of hierarchial structures.

mitya's notebook
A fragment from Pillars of Power. The chromaticism and counterpoint was modeled after Stravinsky's Petrushka.
Robota 2022

A curation of ideas refracted through generative tools: a language model that develops the ideas into longer text, an autoregressive tts to synthesize it, a diffusion model to visualize the text, laid on top of a score generated using evolutionary techniques. The work was an experiment designed to study the confluence of human expression with generative tools.
[ more ]

mossy may
Excerpts from Robota. A Quality-Diversity method was used to generate the distinct voices by using a bespoke fitness function that measures the distance of the voices to the standard created by Johann Fux's counterpoint method.
On Escaping 2021

A profound sense of isolation, darkness, and sombre solitude. The work was a mixed medium installation with a variety of scenes projected on interactive screens that respond both visually and sonically. The viewers were invited to virtually paint over these scenes as they navigated through them. The work seeks to place a splinter in the viewers' eyes, as it were, to act as a magnifying glass that invites listeners to confront and empathize with the intricate web of emotions that accompany the expression of mental turmoil.
The melancholic and dissonant pieces of music composed for this work attempt to mirror the internal conflict and feelings of disquiet, while moments of quiet reflection offer glimpses of fragile resilience and a yearning for resolution.
[ more ]

A fragment from On Escaping. The motivic material was modeled around the fourth movement of Mahler's 9th. The work uses sparse orchestration to evoke fragility and vulnerability.
the process

As a researcher, I feel obliged to provide a rough sketch, at the least, of what this aleatoric and computational process of composition looks like. For this sketch, I used a recording of Bruckner's 9th (Claudio Abbado, Lucerne Festival Orchestra; Deutsche Grammophon) as the source material. The goal is to compose a piece that takes a detour through the Brucknerian landscape by effectively transforming the themes and motifs presented in the Bruckner.

extraits - the orchid in pandyssia

Contour Decomposition

The first step was to model the dynamic shape of the source to preserve its overall contour. This is achieved by first extracting the amplitude data (amplitude over time) from the source, which is essentially what you would see when you open the audio file in an editor. The amplitude over time captures the loudness (dynamics) of the source over time:

Next, Fourier analysis is applied to this dynamic contour to deconstruct the variation of dynamics over time into a combination of oscillations, sine waves, at different rates. Slower oscillations represent gradual motion, while fast oscillations represent rapid changes. The Fourier analysis allows us to take the source material and effectively decompose it to a repertoire of oscillations that when summed together can faithfully recreate the source.

The slow oscillations are used as a guide to create big swells (crescendos) and larger overarching development of the piece whereas the faster oscillations are used to define more energetic shorter spurs and leading voices. These were then orchestrated: assigned to different instrument groups. In an excerpt from extrait's sheet music shown below, notice the violins and violas playing shorter spurs corresponding to the faster oscillations, while the celli and basses start longer expansive swells:

The translation of the oscillations produced from the Fourier analysis to the orchestration ensures that the when then instruments are superimposed (play together by a string ensemble), the dynamic contour of the piece is remarkably similar to the contour of the original Bruckner.

Finally, to maintain a close structural relationship, the counterpoint and voices for extraits were carefully crafted to mimic the overall periodicity, symmetry and cyclic motifs found in the Bruckner. This was an arduous yet rewarding process that imbued extraits with a unique Brucknerian essence. This structural relationship between the two pieces can be visualized via recurrence plots:

Harmonic Landscape - Determinate Negation

While the orchestration of extraits was based on the transformation of the dynamic contour of Bruckner, the harmony was based on a completely different analytical technique: the Tonnetz. The Tonnetz is a lattice representation of the affinities and relationships between notes. More specifically, its a toroidal space of intervals between pitches that can be used to analyze the changing harmonic centers in a piece of music. I designed a particle simulation in which particles are released in a Tonnetz lattice and are attracted to the pitches in the lattice that are currently present in the source audio. The simulation depicts the harmonic landscape of the Bruckner as a function of time (left):

I decided to reverse this harmonic landscape for extraits. Unlike before, the particles released in the Tonnetz lattice are now repulsed by the dominant pitches from the Bruckner (top right). Therefore at any point in extraits, the pitches that are dominant would be the inverse pitches of the Bruckner at that point. In a way, this reversal acts as a Hegelian negation: the very absence of the precise harmonic language of the Bruckner gives extrait's harmony its identity. This creates an interesting relationship between the shared, yet transfigured, harmonic language between the two pieces.

The result of this process is a refraction, effectively a transformation, of the Bruckner and yet sounds nothing like it. Computational and aleatoric/stochastic tools when combined in meaningful ways provide a medium for exploring new ways to articulate rich latent ideas from text, sounds and visual art.