TEK: Listening for Meter and Timbre in Chicago Footwork 

Project Reports
Written By Jeremy Tatar & Victor Burton
 
 
 
 

TEK: Listening for Meter and Timbre in Chicago Footwork 

Interactive Project Report

Published: April 30, 2025

Authors 

Victor Burton (Universität der Künste Berlin), Jeremy Tatar (Carleton College)

DOI

The following essay and derived sonic system were produced by Victor Burton through a collaborative interinstitutional effort with Jeremy Tatar, supported by the Analysis, Creation, and Teaching of Orchestration (ACTOR) project in Montréal. An excerpt of the text, intertwined with the first sonic system, was presented remotely at the Y6 ACTOR Symposium held in summer of 2024 at the University of British Columbia in Vancouver, Canada. A subsequent lecture also paired with the presented system was performed live in Berlin at Cashmere Radio in July 2024.

Throughout the process of writing this text, various conversations were carried with GPT version 4o to explore and clarify technical concepts regarding signal analysis. However, the following text was in no way generated, altered, or edited by the GPT engine. The python scripts developed for the integration of the various wavelet transform operations were conceptualized by the authors and realized by GPT 4o/o-1.

1.0 Preface

 

“Analysis, as it is currently practiced in the United States and Canada, […] enables music theorists to translate musical sound into terms that are transparent, legible, and relatable to Eurocentric knowledge systems.”[1]

 

In her 2002 article “The Paradox of Timbre,” Cornelia Fales proposes a distinction between the acoustic, physical world that sounds inhabit and the perceived world that we ourselves occupy. As Fales writes, “the acoustic world is the physical environment where sound as acoustic signal is produced and dispersed; the perceived world is the subjective, sonic world created by listeners as a result of their translation of signals from the acoustic world. The acoustic world is available through deduction and calculation, but never directly experienced.”[2]

One of the reasons we cannot access the acoustic world is because our perceptual systems—so finely attuned to our surrounding environments—are always-already parsing auditory information according to its source, location, medium of transmission, and so on. This is a snare drum close to me; this is a bassoon at the back of the orchestra; this is my name called out across a crowd; this is a predator across the plain. Fales’s work forces us to confront the idea that these percepts are, in fact, fictions. All that we can ever inhabit is a fictional system of sounds, ideas, and images that is gleaned from an acoustic world outside of our reach. What is listening, what is timbre, what is meter, other than the accumulation of these perceptions?

Fales’s work is deeply inspiring, but there are certain aspects that remain difficult to grasp. What might this acoustic world actually entail? Although we can conceptualize its contours, our subjectivities nonetheless prevent us from accessing it. What might it mean, in other words, to listen with another’s ears? How might we experience a different notion of timbre from the one that our brains—whether we like it or not—force upon us? And how might confronting such a listening open new opportunity for understanding our own musical realities? In what follows, we present a philosophical query into these questions of subjective experience, derive a new methodology for parsing the sounding entities in a recorded track, and close with an extended excerpt from a modular, multi-channel sound work that was created via the resultant computer-assisted analysis. Throughout, we have sought to explore a generative rather than extractive relationship with both our object of analysis as well as the analytical process itself.

1.1. Bangs & Works Vol. 1

Released in 2010 by the English record label Planet Mu, the compilation album Bangs & Works Vol. 1 includes twenty-five Footwork/Juke tracks by fourteen key producers of the genre. In their order of appearance across the album:

 

DJ Elmoe

 

DJ Rashad 

 

Tha Pope

 

DJ Roc

 

RP Boo

 

DJ Spinn

 

DJ Killa E

 

DJ Lil Rome

 

DJ Yung Tellem

 

DJ Trouble

 

Traxman

 

DJ Diamond

 

DJ Nate

DJ Clent                                                                                                                     

 

Bangs & Works Vol. 1 garnered global attention to the underground and highly localized Chicago-based dance music of Footwork, along with its affiliated genres of Juke and Ghetto House, which is characterized by a fast-paced tempo hovering between 150bpm and 160bpm.[3] While every contributor on the compilation had connections with Chicago, they represented different and sometimes overlapping crews, including but not limited to Beatdown House (DJ Clent, RP Boo), TEKK DJZ (Traxman, DJ Killa E, DJ Lil Rome, DJ Yung Tellem, DJ Trouble), Juke Trax (DJ Diamond, DJ Nate) and the Ghettoteknitianz (DJ Rashad, DJ Spinn), the latter primarily known under its new name Teklife since 2011.[4]

1.2 TEK

In his 2021 book Teklife, Ghettoville, Eski: The Sonic Ecologies of Black Music in the Early 21st Century, Dhanveer Singh Brar highlights the ubiquity of the neologism TEK across the Chicago Footwork scene, where it repeatedly surfaces as a “marker of distributed intelligence.”[5] TEK is a sign that is symptomatic of a shared unreality of a racialized urban experience in the economic warfare of a neo-liberal North American megacity. Footwork producers, at times identifying as Teknitianz or Architeks, arise as the actors responsible for shaping both the Teknology and its propulsive potentiality. “Footwork,” then, appears as an engineered tapestry of reactive collisions between high-impact, electron-driven blasts and throbbing subsonic perturbances, and is simultaneously reminiscent of a world that never was and anticipative of one that might never be.

Once channeled under the form of acoustic periodic dislocations and deployed within communal gatherings, the TEK serves as the vehicle for fast-paced, dexterous, gravity-defying dancing. It is as if the sonic manifestation that is heard as Footwork is an audible manifestation from the TEK: Footwork presents itself as an archaeological fabric, in which, interwoven and obfuscated, lie the residual traces of the surging structural forces—the emergences—from/within the TEK.

1.3 Towards Emergence; Whitehead's Nexus Concept

TEK’s contours may be explored through English mathematician Alfred North Whitehead’s speculative philosophy of becoming, as presented in his 1929 book Process and Reality: An Essay in Cosmology. In his book, Whitehead introduces the fundamentals of his “philosophy of organism,” in which reality is in a constant state of becoming.[6] In the first chapter, Whitehead imparts two philosophical warnings: the “fallacy of misplaced concreteness” and a critique of the commonly assumed relationship between logical procedure and certainty. As Whitehead writes, “Philosophy has been haunted by the unfortunate notion that its method is dogmatically to indicate premises which are severally clear, distinct, and certain; and to erect upon those premises a deductive system of thought.”[7] Here, Whitehead cautions against a rationalist paradigm based on immovable categorial schemes aiming to be recast into a “logical truth” about the universal nature of things.[8] Through this, Whitehead further contextualizes his intentions in conceptualizing an organismic ontology of change, in which the essence of his speculative philosophy is to exhibit the abstract character and emergent coherence of reality.

Actual entities are the smallest fragments of reality, sustaining a perpetual process of change. Contrary to an apriorist substance-first ontology, actual entities are characterized by their relatedness rather than the one of primariness.[9] Actual entities (hereafter referred to as AEs) may take the form of any living organisms or material bodies, and are divisible in an indefinite number of ways.[10] Once an AE comes into being, it is no longer an experiencing subject; it perishes and remains as objective datum, for other AEs to form.

 Prehensions, from the Latin prehendere (“to grasp”), are elements of objectification that will be perceived, or prehended, by another AE.[11] AEs are in a continuously renewing relation with each other through prehension. A prehension embodies the uniqueness of an AE that, when interpreted by another, forms a relational link. Three elements are required in the formation of a prehension: (1) subject—the prehending AE (who); (2) objective pole—the analyzed “datum” from the prehended AE (what); and (3) subjective form—the qualitative interpretation of the prehending AE (how).[12] The three constituting factors of a prehension lead to a feeling, not in the emotional sense but precisely as the subjective and determinate way AEs interact and integrate in the world.

 Concrescence arises when a culmination of many interrelated prehensions (and feelings) eventually leads to a “production of novel togetherness.” To concresce corresponds to the process of becoming whole through the integration of a related braid of prehensions, which is the determining role in making an entity “actual.”[13] The final phase of a concrescence, the “satisfaction,” marks the transition from the potential to the actual, from the becoming to the being.

 Nexus (plural nexūs) defines a collection of AEs that exhibit strong relatedness through the interrelation of their prehensions.[14] Unlike the AEs that it comprises, a nexus is not itself carrying the subjective process of prehending. Yet, a nexus itself can be prehended as one overarching pattern of relation. The relational net of AEs and prehensions represented by a given nexus is not an exclusive, closed ecosystem; various nexūs may coexist and overlap. Prehensions of various potencies may form between AEs from dissimilar nexūs. No AE (nor nexus) is truly isolated from the influence of another entity. The nexus, encompassing a system of relationships, subsequently emanates a form of influence back onto the AEs’ relational web, establishing a recursive, tridirectional dynamic between nexus, prehension, and AEs.

Figure 1: A diagrammatic metaphor portraying the nexus of meter within Whitehead's dynamic cosmos: actual entities (AE shapes), prehensions (the forming links), nexūs (dotted lines), concrescences (the radiating process from prehension to AE, ad infinitum).

Thinking of the TEK as a nexus invites speculation as for its AEs and its ensuing system of interactions (prehensions), itself the terrain for emergences to spawn (concrescences). Emergences are embedded in such a way that they impose themselves as boundaryless agents of resistance to understanding. Footwork concresces into audibility from the cumulative mesh of prehensions between multiple AEs: a finessing DJ, a spinning track, an electrical grid, a pulsing loudspeaker membrane, a community center’s linoleum floor, an oscillating air particle, a vibrating cochlea. The being of an AE is constituted by the becoming of its experience.[15]

1.4 Emergent Meter / Emergent Timbre

A filtered clap sounds—and perishes; a synth pad sounds—and perishes; a sculpted kick sounds—and perishes. Once an AE concresces, fulfilling its becoming and coming into being, it perishes. Yet, a perished AE does not vanish into cosmic irrelevancy; its datum remains as a trace of influence for future prehensions and eventual AEs to form. Between AEs of the past, immediate present and future, relationality persists, providing continuity to the experience.[16]

In music, meter initially appears as irreducible and seemingly prohibitive of a rational interpretation through analysis. At the same time, however, meter functions as an equivocal yet ascendant organizational element in the temporal construction of sonic events. Meter is not heard as sound; instead, it constitutes an impression transpiring from the audible sonic construct. Meter emerges from its wholeness. Meter’s ambiguity and elusiveness is why it often receives less direct engagement than its affiliated concepts such as time signature, tempo, phrase, rhythm, pulse and beat, all of which provoke a distinct sentient experience and are commonly evocated in colloquial contexts and subject of perceptual or formal research. Music theorist Richard Cohn points at the long-lasting struggle of attending to meter as residing in the complexity of time itself: “…insubstantial, intangible, unfathomable. Because meter inherits this complexity, it is too difficult to study; either you sense it or you don’t.”[17]

Timbre, too, is also understood as a property that emerges from the interactions of several features of sound, including but not limited to frequency, amplitude, attack/decay envelope, spatialization, and timescale. As noted by Seidenberg et al., timbre is both “a perceptual attribute”—in that it exists largely within “the mind of the listener”—and “a property of fused auditory events.”[18] Timbre, to a great extent, can only be approached obliquely and arises not as a single entity but through the continual interaction of many entities.

1.5 Listening Through Meillassoux's Super-Contingency

As implied by Whitehead’s process philosophy, any attempt to inquire, attend, or listen to the TEK or any of its emergences—such as meter or timbre—invites an epistemological shift that is sovereign from any intent at rationalization, in favour of a frame of thought reliant on the experience of relation.
To form a listening scheme aimed at overcoming correlationism implies a reconsideration of the process of absolutization. Emmanuel Kant’s transcendental idealism presents an avenue through which external objects only appear through our intuition but are merely representations of Ding an sich (thing-in-itself), such as the TEK. The inaccessibility of the thing-in-itself necessitates its own existence along with its own unknowable structure.[19] Thus, in the Kantian sense, the thing-in-itself cannot be known but can be thought through categorization, offering a form of metaphysical realism in which “all subjectivity is correlated with transcendental subjectivity.”[20]

Initially, this form of idealism may appear as an adequate paradigm to prevent any intuitive statements, conjectures, or inferences by the listening subject, knowing that the true nature of things is external to the subject’s perception. However, as the French philosopher Quentin Meillassoux argues in his 2006 book Après la Finitude, relying on a necessary correlation between the listening subject and the unknowable yet thinkable “thing” poses an issue: it implies that knowledge would necessarily be determined in relation to the listening subject.[21] Furthermore, in the pursuit of an adequate listening scheme, this would imply the difficult situation in which any inevitably extractive listening attempt by a subject (be it human or machine), through their own process of reason, would be responsible in determining the TEK. Meillassoux rejects such a problematic epistemic correlation by solely focusing on the very contingency of the correlation. More precisely, as argued by English philosopher Ray Brassier, “the absolute, understood as what is necessary, is not what is but what could be.”[22] There is no necessity for an absolute to depend on a beholding subject. In what Meillassoux calls “supercontingency,” for no or every reason(s), the TEK could be or could have been otherwise at any moment. Thus, as Brassier explains, absolutization of the very fact-ness of the correlation through supercontingency de-absolutizes the correlation itself.[23] This invites a paradigm shift in which the listening act would be speculatively re-absolutized through its contingency instead of its reason, recalcitrant to any further attempt at an eternal conclusion.[24]

Thus, to use a digital methodology to explore Bangs & Works Vol. 1—a methodology of numbers through the aid of the modern supercomputer—is not to be understood as a preaching of numerical supremacy. The technocratic endeavor of listening through the machine is necessarily contingent on its own industrially defined context along with the one of the operator, inevitably antagonistic to any overarching claims of objectivity. Yet, the eventual meaning derived from the methods of science is not necessarily nullified altogether; once repositioned accordingly within the speculative scheme, they hold potentiality as processes of symbolic value. For all one knows, this exercise in speculation may generate a distinct idea of meter and/or timbre.

2.1 NMF (Non-negative Matrix Factorization); Speculating Entities

Through a machine-listening experiment focused on the Bangs & Works Vol. 1 compilation, the presupposed AEs comprising the TEK were examined. Essentially the product of fiction, what follows is the methodology and process of a speculative investigation on meter and timbre as emergences from the TEK.

 NMF, short for non-negative matrix factorization, is an algorithm that decomposes a given dataset into a subset of components. Only prompted by a fixed number of components and otherwise unbiased by any preestablished “knowledge,” the algorithm begins to scan iteratively through a matrix, such as an audio file. It then recognizes recurrent spectral templates and their activations, which are stored in two distinct matrices, and these components are synthesized by factoring the values of its given spectral template with its corresponding activations. The factorization process is then replicated for n number of components. If summed together, the factored components reconstitute the original signal.

Figure 2: Visualization of NMF process, showing the de-composition of a given matrix X into its constituent values for A (spectral template/frequency) and B (activations/temporal dimension).

NMF is characterized by several key attributes. First, the algorithm works in an unsupervised manner. It is automatic and autonomous, and does not rely on additional data to guide its actions.[25] It enacts a form of naive listening, self-determined by its own iterative processes. Second, NMF is non-lossy, meaning that no energy is lost in its application. The sum of its derived components will always be 1, perfectly reconstituting the initial signal. Third, the factorization process of the algorithm is considered “NP-hard”; no exact solution can be determined within polynomial time.[26] Rather, the process is guided by an intrinsic, heuristic-based method that is inherently approximate.[27] Thus, only one thing is certain: that the nature of the decomposition process—and subsequent components—are uncertain.

For this application, NMF “listened” to each of the twenty-five tracks from the Bangs & Works Vol. 1 compilation and stored the results of its iterative process into a matrix. From here, the algorithm decomposed each track into a subset of five components, or “entities.” In some respects, the NMF process resembles the kinds of algorithms that have recently proliferated for stem-splitting. But here, however, rather than being primed and directed to isolate “vocals” or “drums” or “bass” as distinct layers, the resultant NMF entities are voided of any context other than the source track itself. To a human listener’s ear, a given component may resemble an 808 bassline, an R&B vocal sample, or a 909 clap. However, the referential quality of a component remains purely in the realm of the expression. An expression capable of symbolic value, yes, yet one that remains merely and wholly speculative: the audibility of a component does not imply anything as for its conditions. The following six audio excerpts demonstrate this decomposition via NMF: the first is the input track, DJ Elmoe’s “Whea Yo Ghost At, Whea Yo Dead Man,” beginning at 1:21. The five subsequent excerpts are the different resultant entities, each beginning at precisely the same timestamp. In its context-free listening, the NMF isolates and groups salient audio features on a level independent of human-informed hearing.

The implementation of NMF used here was developed by the Fluid Corpus Manipulation (FluCoMa) team at the Huddersfield University, ported by James Bradbury as a lua script usable within Reaper (packaged as ReaComa).[28] Subsequently, per our suggestion, Bradbury generously modified his NMF ReaComa script so as to also save the activations and bases as audio buffers in the project path, as was possible with the original Max version of NMF by FluCoMa.

The resultant activation buffers can be downloaded here.

2.2 Variational Autoencoder; Forming Prehensions(s)

NMF synthesizes its components in no particular order, which prevents any immediate comparative attempt across Bangs & Works Vol. 1. To observe how a given track’s components may relate to those of another track, assessing the spectral template (frequency content) and activations (temporal information) may provide insights on the inner workings of the spontaneous “ear” developed by NMF when applied to a given track. In our application, by considering the equilibrium point in the spectral template of a component (i.e., the spectral centroid), the five components of each track were reordered, from lowest to highest. Through this classification, conjectural links may be drawn by looking at components sharing the same index number, between one and five, across the compilation.

To avoid any human bias in conjecturing these inter-track connections, we attempted to examine such links by using the variational autoencoder RAVE (Realtime Audio Variational autoEncoder), developed by Antoine Caillon at IRCAM.[29] Similar to NMF, the autoencoder “listens” to the input material in an unsupervised manner, unprompted by predefined labels or target outputs. During the training phase, the RAVE model strives to capture the input’s properties in an abstracted way by constructing a series of latent dimensions, which are a compressed representation of the provided audioset’s features. In the present case, the five spectrally defined datasets (i.e., the resultant “entities” from the NMF procedure), each comprising twenty-five components (i.e., one for each track on Bangs & Works Vol. 1), were used as the dataset to train five distinct models. The intent was so that the resulting five models will be “charged” with the timbral limits and spectral dynamics of the respective fictional entities used as training data. For the training, the following attributes were used:

 

rave train --name MODELNAME --out_path ~/ --config v2 --config wasserstein --override PHASE_1_DURATION=1000000 --val_every 5000 --channels 1

 

The training for the five models was achieved remotely on a CUDA-compatible Linux machine located in Montreal via Secure Shell (ssh) access from Berlin. Training of the models was stopped at 1.5 million steps, monitored with Tensorflow. Empirical testing slightly over 1 million steps greatly improved greatly the audible quality of the model, but little gain was observed in going above 1.5 million in this context. The Wasserstein configuration used for training invariably creates a model with sixteen latent dimensions.

These five trained models can be downloaded by accessing this article’s DOI:

DOI

2.3 Continuous Wavelet Transform; Tracing Concrescence(s)

In what is made audible by/from the TEK, time and frequency are inevitably mutually codependent, and thus benefit from a conjoint assessment. The Fourier Transform is often the preferred choice for audio analysis. The Fourier Transform takes an audio signal as an input function from which it can determine its individual frequencies (partials) and corresponding amplitudes, which offers precise frequency representation across the spectrum. However, the Fourier transform is performed across the real axis as whole, omitting any sense of frequency localization in time.[30] In order to obtain the temporal dimensions of the signal, the Fourier transform is performed periodically over several short-time windows, yielding what is known as the Short-Time Fourier Transform (STFT). When dealing with the STFT, there is thus a case-specific negotiation between larger time windows, which favor precise frequency resolution at the detriment of the temporal information, and smaller time windows, which inversely favor precise temporal information at the compromise of the frequency resolution.[31]

An alternative signal analysis method is the Continuous Wavelet Transform, which offers a solution of sorts to the dilemma of the STFT. The core procedure in the CWT involves a “wavelet”—a brief oscillation with beginning and ending points valued at zero—which is continuously shifted across the signal. Akin to a stencil, the wavelet scans along the time plane and stores the value corresponding the signal’s phase alignment with the wavelet. The process is then repeated with the same wavelet type yet scaled differently, where the wavelet, the listening stencil, is compressed and expanded.

For each scale of the wavelet’s compression/expansion, on a per-sample basis, the wavelet information is stored under a pair of coefficients in the two-dimensional complex plane. On a short segment of all the 125 components suggested by NMF, the wavelet transform was performed with sixteen scales, corresponding to the number of latent dimensions in each of the models. The two axes of the complex plane (the real axis and imaginary axis) conjointly express the wavelet’s phase and amplitude at a given scale. Thus, the complex plane portrays a time-frequency—or rather time-scale—representation of the signal, embedded in the repartition of its values.

Figure 3: Visualizations of one-measure segments from two components extracted from DJ Clent, “I Love You”: Entity 3 (left) and Entity 5 (right).

Figure 4: Two-dimensional visualizations of the same two excerpts, showing only the complex plane.


Then, for each scale, the average skewness of both the real axis and imaginary axis was calculated. In this context, skewness expresses the slope towards which each axis of the complex is tending. Abstruse and derivative, as if arising from the manifestation of some influent concrescence through unknown prehensions, skewness as a time-scale tendency information may offer an abstract representation of an underlying metric force, or feeling, a hypothetical signifier of meter.

Figure 5: Skewness calculations for excerpts from DJ Clent, “I Love You,” showing both the real and imaginary axes.

This resulted in a rather large array of sixteen skewness values (derived from both the real and imaginary coefficients of eight different scales) for each of the 125 components. Based on the skewness value of the real axis for the first scale of each of the 125 arrays, the arrays of skewness values were then reordered from lowest to highest.

In order to interpolate between the values, a lerp function was used within Max using the javascript (js) object:

 

var data=[ 

[-1.9135600931070105,-1.9462285825460033,-0.46713428759254577, ...etc. ]

[

function lerp(b,a,t) {

        return a *t + b * (1-t)

}

function listlerp(la, lb, t) {

        var value =  new Array();

        for (var i = 0; i < data[la].length ; i++) {

                value.push(lerp(data[la][i], data[lb][i], t))

        }

        outlet(0, value) 

}

function scan(s) {

        var b = Math.floor(s) // entier (valeur de gauche)

        var r = s % 1.0 // "restant"; valeur d'interp

        listlerp(b, b+1, r)

}

The “position” within the complex plane can then be incremented between the skewness values of two analyzed components. Then, with a midi or OSC controller, one can continuously cycle through skewness values that would be “in between” certain components. The data is used as a modulation factor for each of the sixteen latent dimensions of the model (see below). The lerp function, containing the skewness values described above, can be downloaded here.

2.4 Playing the Sonic System

Finally, the resultant data from the NMF listening and RAVE autoencoder were mobilised for creative purposes as part of the research-creation component of this project. RAVE is generally approached with the intent of performing timbre transfer, in which a source sound will be analyzed in real time, encoded into a relevant control stream for the latent dimensions, and then decoded by the model. To the best of the model’s ability, it will mimic the source sound’s behaviours yet enact its own timbre. Alternatively, the model can also be interfaced with by skipping the encoding process and instead directly interfacing with the latent dimensions. Since the RAVE variational autoencoder weaves its latent space as a distributed representation in which the analyzed data is non-linearly entangled, the size of each latent dimension and the direct impact of its output on the sound resynthesis cannot be predicted.[32] Thus, there is no way deemed “appropriate” to interface with the model’s latent dimensions. Any approach shall remain in the realm of the symbolic.

The system is put into action by using the activations buffers used by NMF at the very first step of the process. Akin to envelope followers, they are used as control signals sent to the RAVE/nn~ decoder object. Five group objects (one for each model trained as part of Section 2.2) are responsible of playing back the activations. The activation buffer can be chosen via a umenu object in Max. The parameters for the rate and loop position are mapped on a Twister Fighter MIDI controller, which can be changed dynamically. The complex plane, with its varied bipolar values, is factoring the signal coming from the activations before being transmitted to the model. This way, it behaves as a control data “filter” that splits the mono activation buffer into sixteen distinct channels that will then be sent to the latent space, resulting in a manifold of timbral structures emerging from complex plane. 

Figure 6: Screenshot of one of the five instances of the Max MSP resynthesis system. PINK: the activation player. The last parameter of the scale object can be used a signal shaper (exp/log curve). The dropdown umenu object at the top chooses the activation. BLUE: the complex plane parameter with a value smoother (for MIDI value jumps) going to the lerp. The list with the 125 by 16 skewness values to be interpolated is stored inside the lerp.js object. YELLOW: unpacks the 16-value list from the lerp. GREEN: the parameter (“1.” here) is akin to a meta “gain” for the control stream, before going into the decoder. The “latent polarity” can be inverted with negative values (ex. -1).

This Max patch can be downloaded here—please note, however, that this is not intended as a streamlined instrument to be “played” or performed; rather, it is one aspect of this project’s open-source documentation, and we invite readers with knowledge of Max and RAVE to explore its workings and potential for further expansion.

2.5 Sonic System: Live Excerpts

This is an audio excerpt from a live presentation of the Max resynthesis sonic system—interacted with via a Twister Fighter MIDI controller—that took place in Berlin at Cashmere Radio in July 2024.

2.6 Reflections on the System

The realization of the sonic system proved to be an insightful process that invited a critical assessment of various signal processing methods which prompted an unconventional implementation of dissimilar algorithms. NMF (for actual entities), RAVE/nn~ (for prehensions), Continuous Wavelet Transform (for concrescences) each played a role in the symbolic re-contextualization of the original audio material from the Bangs & Works Vol. 1 compilation, inspired by Whitehead’s philosophy.

Following two public presentations of the system, in August 2024 it was understood through discussion with Antoine Caillon that RAVE uses a “compression” ratio in its latent space. Any data stream sent directly to the nn~ decoder is subjected to a form of chunk buffering, in which the time resolution of continuous data stream is downsampled. This compression value is fixed: 2048 samples at 44.4kHz, which means that RAVE has a “refresh rate” of 44100/2048=21.53Hz, equivalent to approximately 21 “updates” per second. The most significant downside of this constraint is that any periodic control signal that is sent to the latent dimensions of the nn~ decoder (such as a simple linear phasor, or, in the present case, the activation buffers) will get “latched” and become jagged due to the constant compression ratio. The imposed compression ratio limits RAVE in its ability to precisely interpret any timely pattern; the impact is salient both rhythmically and timbrally. This property is dependent on the rate of the given controlling signal in relation to the fixed value of 21.53Hz. Thus, a hypothetically stable, periodic activation such as linear control phasor with a speed of 3Hz will succumb to a different temporal and timbral distortion than a linear control phasor at 4Hz. This inevitable artefact results in the model being considerably unpredictable and uneven in the timeliness of its sound production, regardless of the nature of the controlling data signal going to its latent dimensions.

3.1 Conclusion

Bangs & Works Vol. 1 features in this project as a set of significant tracks by a curated set of Footwork producers. Yet it is also imperative to dedicate avid listening to the rest of the artists’ releases, as well as to the other sonic outputs originating from the city of Chicago. The nexūs of meter and timbre emerges not as static entities but as dynamic and interconnected patternings of becoming. In listening through the ears of NMF and the CWT, we approach an unmediated encounter with the AEs that constitute this particular musical form. Crucially, the sonic system presented in this study stands as but one concrete instantiation of the abstract form of the nexūs of meter and timbre, reinforcing a plurality of possible conduits for experience. Procedural, the TEK of Footwork exists endlessly in the potential of what it could be.

Works Cited

Jeremy Tatar & Victor Burton
Previous

Orchestration as Form in Elisabeth Lutyens’s And Suddenly It’s Evening (1966) 
Next

The Music of Dialect