Broken on purpose
From the torn speaker cone on Rocket 88 to latent-space vocal models: how audio's accidents became methods, and whether building the break on purpose keeps it alive or kills it.
The story of the first distorted guitar is told several ways, which is the first thing worth noticing about it. In one telling, the amplifier fell from the trunk of a car on the drive into Memphis; in another, the speaker cone was soaked through by a sudden rainstorm; in the version most often repeated, the cone was simply torn in transit and packed with wadded newspaper to stop it rattling itself apart. What the tellings agree on is the year (1951), the place (Sam Phillips's recording studio), and the outcome: when Willie Kizart plugged his damaged amp in to cut "Rocket 88" with Ike Turner's Kings of Rhythm, the sound that came out was not the sound of a working amplifier. It was a fuzzed, broken-up texture, the audible signature of a thing failing to do its job. Phillips, who might reasonably have sent out for a replacement, kept the take instead. He built the record around the buzz of a dying speaker, and a sound nobody had designed became a blueprint. Whether or not the details are exact (and the fact that the anecdote survives in irreconcilable versions suggests it has become a myth, a structure repeated for utility rather than historical record), the shape of the event is perfectly legible. A broken thing is kept on purpose. Someone notices the break, and decides the failure is better than the function. The question this lineage actually pursues is what happens when noticing becomes a method.
The Roland Corporation spent the early 1980s chasing the opposite of an accident, building machines to execute a clean, reliable job (specifically, to stand in for session musicians), and by their own metrics, the machines failed. The TB-303, sold as an automatic bass accompanist, was dismissed for the thinness of its synthetic tone; the TR-808 made percussion too obviously electronic to pass for a drummer. Both were commercial disappointments against the parameters they were designed to meet. Yet within a few years, producers in Chicago's underground were torturing the 303's filter into the liquid squelch that would name acid house, and the 808's unnaturally deep, tunable kick was settling in as the foundational low-end of hip-hop. The same escalation reached software by the end of the century when Antares Auto-Tune (released in 1997 as an invisible corrective for pitchy vocals) was pushed so hard on Cher's "Believe" that the correction stopped sounding like a subtle nudge and started sounding like a machine breathing through a person. What matters here is the velocity of the reappropriation. These tools were not left to slowly decay on a shelf for decades to acquire a melancholic patina. The drum machines were weaponised within years, not decades, and the pitch correction within months. This is the first structural escalation: from the accident found, to the accident exploited on purpose.
Consequently, there is a fork in the taxonomy here. One route involves the slow degradation of physical media (the hauntological decay of William Basinski's rotting tape loops, a lineage of mourning), and it is a route this essay is explicitly leaving behind. The other route turns from repurposing a failure to actively seeking the wreck out. When the French urbanist Paul Virilio observed that the invention of the ship is simultaneously the invention of the shipwreck, he established the integral accident as a structural guarantee of any new technology: every system ships with its own specific disaster baked into its architecture. But the history of audio hardware suggests a necessary extension to Virilio's law: the shipwreck inevitably invents its own devotees, and eventually, its own architects. The practitioner emerges who genuinely prefers the failure to the success.
If you prefer the break, the logical next step is to stop waiting for chance to deliver it. (This preference is not an abstract theoretical condition for me; it is a methodology I have spent the last few years chasing through custom soft-synths, a half-hobbyist obsession with training acoustic models on my own voice not to cleanly replicate its timbre, but to isolate the exact parametric threshold where the latent space tears). In 1966, a teenage Reed Ghazala accidentally short-circuited a battery-powered toy amplifier against a metal desk drawer, producing a torrent of synthesised noise. Rather than fixing the short, he began deliberately wiring switches into the logic boards of consumer electronics to trigger these exact cross-circuits on demand. This practice (which Ghazala would later formalise as circuit bending) acts as the hinge. It represents the transition from luck to discipline. The torn speaker cone in Memphis was a statistical anomaly; short-circuiting a Speak & Spell on purpose to map its undocumented screaming is a method. The consequence of this shift is profound: once a threshold can be triggered, it can be mapped, and if you can induce failure systematically, you can systematise the failure itself.
The digital equivalent of the induced break arrived with the optical disc. While audiophiles celebrated the sterile, noise-free perfection of the compact disc, artists like Yasunao Tone began deliberately wounding the media with pinpricks and tape, forcing the laser to stutter and violently misread the data. The German electronic project Oval adopted a similar approach for Systemisch (1994) and 94diskont (1995), painting the underside of CDs to harvest the skipping audio buffers. This practice of digital damage received its formal manifesto in 2000 with Kim Cascone's The Aesthetics of Failure, the exact moment the quirk was named as a programmatic movement rather than a fleeting error. Cascone's foundational text carries particular weight because he was not merely observing from the academy. He had founded the ambient label Silent Records in 1986, and worked as an assistant music editor for David Lynch on both Twin Peaks and Wild at Heart, absorbing an industrial-surrealist audio environment where ambient dread was treated as a physical material, before systematising that underlying logic into a theoretical framework a decade later. The practitioner, having touched the metal, always precedes the theorist.
But scratching a CD still relies on a physical vulnerability, a retroactive attack on a finished object. Markus Popp of Oval understood that exploiting a broken disc was still fundamentally an act of harvesting. The definitive move the entire history has been climbing toward since Memphis is the transition from finding the break to building the machine that makes it on demand. Popp commissioned Ovalprocess, a custom software environment that algorithmically systematised his own CD-skipping methodology, generating the exact acoustic properties of a damaged CD player as a controllable parameter. The engineered artifact is the work, and the software is simply the factory built to produce it reliably. There is an unsettled friction here, however, that the lineage rarely admits. The broken amplifier in Memphis was beautiful partly because it was an unwilled, unrepeatable accident. The moment you can produce the glitch perfectly every time, you have domesticated the wildness that made it worth wanting. The failure ceases to be a break and becomes a preset. Is the instrument-builder completing the tradition of the accident, or acting as its taxidermist? The lineage presses forward without resolving the question, haunted by the possibility that in reliably manufacturing the break, it might kill the exact unreliability it seeks.
Pushing this impulse further requires a categorical shift. Granular synthesis (theorised mathematically by Dennis Gabor as acoustic quanta in 1947, mapped compositionally by Iannis Xenakis, and brought into real-time computer music by Curtis Roads and Barry Truax) does not actually belong to the lineage of malfunctions. It is a legitimate, cleanly theorised compositional technique. But it acts as the structural climax of this history because it pivots from imposing a failure onto a signal, to exposing a break intrinsic to the medium itself. Here, the practice stops adding damage to the continuous wave and reveals instead that the continuous wave was always an illusion. Granular processing shatters sound into the microscopic buffer-fragments the medium is secretly made of. The wave is atomised into a cloud of independent grains, each lasting only milliseconds, hovering in a suspended density where time and pitch are decoupled and the architecture of sound is reduced to a shimmering, textural dust. Granular synthesis is not a failure applied to the surface. It is the mathematical exposure of the digital substrate, proving that the signal was always failing to be continuous.
This ontological shift clarifies exactly how to classify the newest medium. Generative AI models ship with their own integral accidents (hallucinated geometries, temporal smearing, the specific woozy warble of compressed latent audio), but routing these through a framework of vanishing bugs is a categorical error. Developers have always tried to patch out the failure. The toolmaker's job has always been to build a system that retains it. Consequently, the artists doing the structurally significant work in this space are not waiting for the system to perfect itself, nor are they mourning the lost artifacts of deprecated versions. Practitioners like Holly Herndon and Mat Dryhurst (with vocal models like Holly+ and Spawn) do not view AI as a black box of unpredictable errors to be smoothed over. They cultivate the machine-strangeness as a native material property. They are the modern luthiers architecting the latent space, continuous with Ghazala and Popp, cultivating the exact strangeness they require and building the tooling to play it. Artificial intelligence is simply the new granular buffer; its artifacts are not glitches to be solved, but the exposure of its own latent substrate.
The pursuit of perfect fidelity and the pursuit of beautiful failure (the developer relentlessly patching the bug on one side, the systems engineer tuning the break on the other) are locked in a permanent antagonism. They are the two halves of Virilio's equation, pulling in opposite directions. The toolmaker who builds the break on purpose has chosen a side, though it is not yet clear whether choosing it is an act of preservation or of embalming. Every invention invents its disaster, and the disaster invents the instrument-builders who learn to play it; what no one can say in advance is whether a failure that arrives reliably, on command, every time, is still the thing they fell in love with. That is the open question the next medium inherits. Its beautiful failure is already coming, and someone is already building the machine to summon it. The only thing left undecided is whether summoning it that faithfully is how you keep it alive or how you finally kill it.