Setup

Amplification & No-Input Mixer Configuration

The cymbal is amplified with a condenser mic (Rode NT5) into Channel 2, summed alongside the mixer's own no-input channels so the acoustic sound and the feedback loop share the same signal path.

Gain & EQ

Gain range depends on input type: 0dB to +60dB using XLR, using Line there is a 15dB attenuation up to 45dB at the knob fully up; the stereo channel runs from off +10dBV to +4dBU (no knobs).

Control techniques

Instrument techniques

The systematic recordings below cover brushes, medium mallets, soft mallets, drum sticks, bowing, and the ribbed-mallet/guiro technique. Nail rubbing wasn't given its own dedicated take, it appears instead inside the full improvisation, alongside guiro-style scraping, as one of the bright, noisy passages discussed there.

Analysis method

Every recording below was processed the same way: onset detection (spectral-flux), spectral-feature extraction (centroid, bandwidth, flatness, RMS) on a 2048-sample STFT, segmentation with clustering into acoustic classes and spectral-peak tracking to find recurring "attractor" frequencies, and isolated-onset case studies to see what the feedback loop does once excited. Two recordings (guiro, bow) use continuous rather than discrete excitation, so their analysis leans more on the overall energy trend across the take than on onset-triggered decay curves.

General Overview & Synthesis

The cymbal and the no-input mixer

The session was structured in two stages: first, systematic gesture-by-gesture studies isolating one mallet or technique at a time; then a full improvisation combining everything into a single continuous performance. Total: 37 minutes 40 seconds, 5,137 detected onsets, eight recordings, six excitation types.

1. The instrument's resonant fingerprint

While the cymbal has a very wide spectrum, this specific no-input mixer setup keeps returning to the same handful of resonant frequencies, almost regardless of what is exciting the cymbal. Tracking the strongest spectral peak in every analysis frame across all eight recordings shows six frequencies that recur, sometimes dominant and sometimes secondary, in every single recording: 110Hz, 185Hz, 320Hz, 400Hz, 585Hz, 820Hz.

Every recording shows one of the six bands, and each band is the strongest lock-in in at least one recording: 110Hz emerges in Soft Mallets Pt.2, 185Hz emerges in Bow, 320Hz emerges in Drumsticks, 400Hz emerges in Medium Mallets, 585Hz emerges Guiro, and 820Hz emerges in Brushes. The frequencies aren't in a harmonic series (the ratios between neighbours run roughly 1.2–1.8), arguing for separate, independent resonant modes of the mixer's system itself, which show up whether the cymbal is bowed, brushed, struck, or scraped.

2. A typology of feedback behaviours

Across all the isolated-decay case studies documented below, the same handful of qualitatively distinct behaviours keep recurring:

No recording shows only one of these; every entry has at least two or three. The two continuous-excitation recordings (guiro, bow) are the only ones to show the cumulative-buildup pattern: a struck gesture delivers one burst of energy and leaves the feedback to do whatever it does next, while continuous friction keeps re-supplying energy at a steady rate.

3. How technique and material shape the interaction

Discrete strikes vs. continuous excitation: This is a structural split in this session, as it directly affects the degree of interdependence of the instrument and the mixer's signals.

Mallet hardness tracks brightness and noisiness: Aligning with the physical intuition about hardness and the contact area of the exciters used in the session the median centroid comparations shows how Soft Mallets (2,067–3,213Hz) < Drum Sticks ≈ Guiro (3,008–3,126Hz) < Medium Mallets ≈ Bow (4,485–4,568Hz) < Brushes (5,810Hz).

Rolls read as tonal: A roll on an untreated cymbal is dense and broadband; a roll into this mixer instead shows very low spectral flatness alongside a high onset rate, meaning the mixer picks a pitch and holds it even while being re-triggered several times a second.

Bell vs. side location produces a detectable signature: Two independent results, a timbral split and a drifting locked frequency, agree on the same boundary.

4. Overall picture

The mixer's feedback path behaves as a nonlinear system with a small, fixed set of preferred resonant states (the six frequencies above) that are largely produced by the electronic system rather than the cymbal. What the acoustic input actually controls, through its wide spectrum, is which of those states gets excited, how strongly, and via which qualitative behaviour. A single quiet, barely audible touch is enough to trigger a full-spectrum buildup that outgrows the gesture that started it. A strong, sudden attack can just as easily produce a clean decay that never catches at all. The relationship between gesture and outcome becomes closer to a threshold system, where the acoustic input's job is to nudge the loop across, or fail to cross, the boundary into self-sustaining behaviour, and once across, the mixer's own dynamics largely determine what happens next.

01 — Brushes

Source: 01_sistematicgestureBrushes.wav · 3:52.77 · brushing, hitting, and rubbing the cymbal

497 onsets over 232.8s, unevenly distributed: near zero through long ringing stretches, spiking above 20s. Clustering the take's gesture segments recovers three acoustic classes, mapped to the three named techniques:

Pitch content: the acoustic brush/cymbal contact itself has no fixed pitch, but once inside the mixer a small set of frequencies recur far more than chance would predict. 823Hz (roughly G♯5) is the dominant lock, present continuously nearly the entire hitting/rubbing portion of the take. 400Hz, 325Hz, 135Hz, and 113Hz surface mainly towards the end of the recording as the loop's resonance steps downward toward silence. None of these sit in a simple harmonic relationship, they represent independent resonant modes of the mixer's own feedback system.

Interactions

Four isolated onsets, each followed by 15+ seconds with no further physical contact, sketch a small typology of what the loop does once excited by the instrument:

02 — Medium Mallets

Source: 01_sistematicGestureMediumMallets.wav · 2:33.15 · rolls, strokes, and hits (strokes and hits each tested on bell and side)

376 onsets over 153.2s (mean 2.46/s), essentially continuous apart from a lead-in and a tail. A roll passage (70.5–84.7s) stands out immediately: onset rate 12.7–23.6/s, yet flatness stays very low: a second confirmation that a fast, periodic gesture interaction with the the mixer render as quasi-tonal sound material rather than noisy. The remaining material splits into a sparser, more focused "strokes" class (4.2–41.0s) and a louder "hits" class (41.0–70.5s, 87.0–147.8s) that drifts in timbre (narrower/darker before ≈112–117s, brighter/broader after) and "cuts" the self-oscillation.

Pitch content: the dominant recurring frequency is 401Hz (roughly G4), locked continuously during the strokes phase, the first hits phase, and the rolls. 576Hz runs even longer, alongside 319Hz and 518Hz over similar stretches. In the window the dominant lock drifts upward through 576Hz, up to 900–1,100Hz, back to 650Hz, settling near 903Hz (roughly A#5) for the final ~10 seconds. Something physical changed in that stretch, correlating with timbral shift.

Interactions

Four onsets, each window cut exactly at the next detected onset so nothing shown is contaminated by a new strike:

03 — Soft Mallets

Two recordings from the same soft-mallet session. Same gesture set as Medium Mallets (rolls, strokes, hits at bell and side)

Part 1 — 6:16.16

329 onsets, concentrated unevenly. Soft mallets produce a noticeably darker and more tonal than medium mallet type.

The take splits into four blocks: an opening, dense/bright block with one sharp roll insert (strokes → rolls), a long, sparse, dark block (hits on bell), and a very sparse, brighter final block (hits on side), similar to the behaviour observed in the Medium Mallets rec.

Pitch content: 319Hz (roughly D♯4) and 401Hz (roughly G4) are the two dominant locks, and both survive between blocks, present early on and reappearing after the loop has gone completely silent. Again an evidence that these are persistent resonant states of the mixer's circuit itself, potentially triggered. A third, quieter 109Hz (roughly A2) lock builds up in the later, darker part of the second block.

No spectrogram/feature-tracking images were generated for Part 1. The four gesture recordings below, are taken from the analysis if the full set.

Interactions — Part 1

Part 2 — 3:56.17

997 onsets and markedly darker still: median centroid 2,067Hz, flatness close to zero almost everywhere. Structure splits into a sparse dark opening, a hits/strokes class, and three separate roll bursts. Combined with the roll instances already documented, the roll signature now counts six independent instances of the "fast-but-tonal" nature across four recordings.

Pitch content is the strongest lock in the session: 109Hz (roughly A2) dominates almost completely: the strongest peak in nearly half of all analysed frames, present continuously from 0s to 233s regardless of gesture. This is the same frequency that appeared only as a secondary lock late in Part 1, and now confirmed as the dominant resonance of an entirely separate take, again a sign of the mixer own self-sutained resonances.

Interactions — Part 2

Cross-recording analysis: Part 2 is darker and more heavily locked than Part 1; possibly becasue of a more consistent soft, central mallet contact throughout. Both parts show the same qualitative repertoire though: regenerative buildup from a quiet touch, extended near-flat sustain, and synchronized whole-spectrum dip/recovery.

04 — Drum Sticks

Source: 01_sitematicgesturesSticks.wav · 3:36.78 · rolls, strokes, and hits (strokes and hits each tested on bell and side)

216.8s, continuous apart from a lead-in and short tail. 559 onsets (mean 2.58/s). Median centroid and flatness lay almost exactly between medium and soft mallets in the "brightness ranking". Clustering finds a clean five-part structure: strokes (bell+side interleaved), a hits region with low onset rate (side), a darker hits region (bell), a roll, and a resumed brighter hits region (side). The bell/side duality rests on the same brighter-and-broader vs. darker-and-narrower contrast found in Medium Mallets.

Pitch content confirms the system's "preferencies": 319–331Hz is the dominant lock here, present continuously from 14s to 213s, with 401Hz alongside it from 19–213s and a quieter 109–121Hz lock from 26–212s. These are the same two dominant locks found in both Medium Mallets and Soft Mallets Pt.1, and 109Hz is the same frequency that dominated Soft Mallets Pt.2. Three completely different excitation materials: soft, medium, and hard stick, and the mixer keeps returning to the same handful of resonant frequencies regardless.

Interactions

05 — Guiro (Stick Rubbed on Cymbal Side)

Source: 01_1_sisteamticgesture_Guiro.wav · 3:02.10 · extended technique (continuous friction)

Median centroid and flatness are close to the Drum Sticks recording (the same stick material, used differently). The defining feature here isn't per-gesture decay but a cumulative buildup: mean RMS energy rises roughly eightfold, near-monotonically, across consecutive 30-second windows over the full three minutes. The discrete-strike recordings doens't behaves like this: continuous friction cumulatively feeds the loop, with later continuation of the gesture needing less effort to sustain a loud, dense texture than the earlier did.

Pitch content: 576–606Hz, 500Hz and 821Hz cover almost the entire take. 821Hz is the same lock (within a few Hz) that dominates both the Brushes and Medium Mallets recordings; 576Hz is likewise close to a lock seen in both. Again, rather than each technique creating its own unrelated set of resonances, the different excitation styles seem to preferentially wake up different subsets of a shared pool of resonant modes of the mixer itself.

Interactions

Four gap-truncated isolated onsets, chosen from the moments where the continuous rubbing did pause:

06 — Bow (Double Bass Bow)

Source: 01_1_sistematicgestureBow.wav · 4:40.42 · extended technique — continuous excitation

Bowing produces a strongly tonal excitation. This take shows two phases: a long buildup across the first ~200s (energy generally climbing instead of resetting after each stroke), then a marked brightening around 200–260s where centroid jumps to a higher range, representation of a change in bow contact point and pressure.

Pitch content: 178–190Hz (roughly F3/F♯3) is present throughout the entire take. Short locks appear in sequence beneath it: 249Hz, 354Hz, 390Hz, and 600Hz. Continuous excitation seems to let the feedback settle into and hold a resonance more than a struck excitation does.

Interactions

00 — Final Improvisation as Synthesis

9:42 · 1,125 onsets · median centroid 4,011Hz · mixed: mallets, rolls, guiro effect, nail-scrape, bow

The full improvisation is what happens when all six are combined with the free-form negotiation between player and feedback loop that the systematic studies were designed to isolate from. This adds to the studies of the isolated gesture, an analysis of transitory parts: what happens when the feedback and the system, left in one state by one gesture, is immediately given a different kind of input by the next.

Interactions

Setup

Amplification & No-Input Mixer Configuration

The guitar is captured with the signal passed through a Helix multi-fx processor before reaching the mixer, so the "instrument" feeding the loop is the guitar plus its own pre-processing chain. Additionally, a contact mic and a condenser mic (Rode NT5) have been used for some of the more experimental techniques.

Mackie 1202-VLZ PRO

Gain & EQ

Gain range depends on input type: 0dB to +60dB using XLR, using Line there is a 15dB attenuation up to 45dB at the knob fully up; the stereo channel runs from off +10dBV to +4dBU (no knobs).

Instrument techniques

Processing before the mixer

Alongside the physical techniques above, the performer also made use of two pre-mixer effects:

Control techniques

Analysis method

Each technique instance below is a short excerpt (6–20 seconds) cut from one of the six raw improvisations. Every excerpt is processed the same way: a frequency spectrogram (2048-sample STFT, 512-sample hop, 32kHz sample rate) and a 4-band energy trend (80–200Hz, 200–500Hz, 1–2kHz, 4–8kHz), to characterise whether the mixer's feedback loop builds, decays, or holds steady during and immediately after the gesture.

The recordings are natively 48kHz, but analysis is at 32kHz (16kHz audible bandwidth) rather than run at full native rate: above roughly 18kHz the recordings carry a persistent, near-flat noise band that doesn't move with the playing and shows up identically in quiet passages regardless of technique. Almost certainly interference from the recording chain itself or my audio interface. Left in, it dominates the brightness statistics for any quiet excerpt therefore I opted for a 16kHz ceiling.

General Overview & Synthesis

The electric guitar and the no-input mixer

The six raw recordings total 1h 47m of continuous free improvisation; fifteen short excerpts have been pulled from them for this document, spanning six techniques.

1. The instrument fingerprint

In this session, the structural split is again (see Cymbal) between techniques that supply energy continuously and those that deliver it in discrete pulses. Bowing and brushing with a contact mic, two of the most continuous ones, produce three of the session's four clearest full-spectrum-buildup or slow-climb cases. The discrete techniques (chords, harmonics, and tapping) show a much more mixed picture: some instances build (plucked and strummed chords), others decay (arpeggiated chords) or settle into a narrowband sustain (harmonics) instead.

2. A typology of feedback behaviours

Across the fifteen marked instances, the same handful of qualitatively distinct behaviours recur, echoing the typology found in the cymbal session:

3. How techniques and material shape the interaction

Attack style decides the outcome: chords played three ways on the same strings span nearly the whole brightness range of the session. Plucked with attack softened by the volume knob, swells into a frequency lock; Strummed with a pick generates stromng burst of energy follwed by continuous distortion and frequency locks; Arpeggiated reaches the brighter and thinner frequency bands.

Pitched content decides what the loop locks onto: unlike the cymbal, whose mixer returns to the same six resonant frequencies no matter what struck it, the guitar's harmonics show the loop locking onto whatever pitch is actually played.

Some continuous techniques transfer better than others: a bow, produces the same clean full-spectrum buildup found in the cymbal's own continuous techniques. A contact mic brushed across the strings is far less predictable.

The string's own decay competes with how fast it's re-excited: a plucked string starts decaying the moment contact ends, so whether the loop settles depends on whether the next attack outruns that decay. A single harmonic touch lets the string's own slow decay carry the mixer into a steady hold, while the tapping pick re-excites it so fast that it never gets the chance to do so.

4. Overall picture

The guitar's harmonic nature changes what the mixer answers to. Where the cymbal's loop falls back on a small, fixed set of resonances regardless of technique, here it tracks the actual pitch and rhythm of whatever excites the string, whether that's a finger, a pick, a bow, or a contact mic. Attack style and re-excitation speed decide whether the loop settles, builds, or locks onto a note. As with the cymbal, this behaves like a threshold system, except here the threshold responds to how a note was set moving rather than to a fixed property of the mixer itself.

01 — Chords

Sources: S02_fourthImprov.wav, S02_firstImprov.wav, S02_secondImprov.wav · plucked, strummed, and arpeggiated chords

The three ways of setting a chord moving produce three different outcomes from the mixer. Plucked with fingers, with the attack softened by the instrument's volume knob: swells smoothly rather than striking, the low and low-mid bands climb for about three seconds and then hold almost flat for the rest of the excerpt, while the top end fades quietly underneath. Strummed with the pick behaves differently: five or six strums are each visible as a sharp broadband transient that decays away in the top band within a couple of seconds, but the 80–200Hz band barely moves across the whole excerpt, holding a near-constant level from strum to strum regardless of what happens above it (mixer locking in). Arpeggiated gives the cleanest decay of the three: the passage excited the high end of the feedback loop, resulting in a grainy high texture, and a 6,812Hz median centroid.

Interactions with the NIM

02 — Harmonics

Sources: S02_fifthImprov.wav, S02_thirdImprov.wav, S02_fourthImprov.wav · natural harmonics touched off the string

All three harmonics' excerpts share a visible, narrow pitched character, but they diverge in what happens to that pitch over time. The first is a single harmonic around 592Hz with only a very gradual decay across every band. The second shows a run of stepped harmonic touches before settling onto one sustained tone near 330Hz while the low end loses energy. The third pushes that same pattern furthest: the bass drops a full 12.1dB across the window while the 200–500Hz band directly above it barely moves (−0.7dB).

Interactions with the NIM

03 — Tapping Pick

Sources: S02_secondImprov.wav, S02_fourthImprov.wav · the pick tapped against the string, one instance a glissando

The glissando excerpt opens with a clean descending pitch line, sliding from around 4kHz down to 1kHz over the first four to five seconds, before breaking into a dense, irregular pattern that doesn't evolve into a steady state: every band stays within a few decibels of its starting level for the rest of the excerpt (the largest drift is only −2.6dB), the tapping re-excites the system too fast for it to either decay away or build up. The second excerpt is much shorter and punchier, in which the feedback visibly begins to capture the low register.

Interactions with the NIM

04 — Brushing with Contact Mic

Sources: S02_firstImprov.wav, S02_secondImprov.wav, S02_fifthImprov.wav · contact mic brushed against the strings/body

The three brushing excerpts show what continuous friction can do to the loop. The first is a continuous gain: every one of the four bands gains a little energy across the full 20.5 seconds (+1.5 to +3.0dB), building a lock. The second is close to the opposite: a thin, scratchy burst with pitch trails runs for about 4.5 seconds before crashing 6–15dB across every band into near-silence. The third is a striking diagonal sweep climbing from around 130Hz past 8kHz through the middle of the window cuts the broadband hiss before the texture returns to noise.

Interactions with the NIM

05 — Bowing

Sources: S02_thirdImprov.wav, S02_fourthImprov.wav · the string excited with a violin bow

Both bowing excerpts show continuous excitation feeding the feedback, but to different degrees. The first, with quick bowing, builds a dense harmonic ladder of stacked partials across the recording; the low end gains 3.6dB while the upper-mid gently recedes. The second, with slow bowing, is a full-spectrum buildup: all four bands gain energy simultaneously.

Interactions with the NIM

06 — Processing Before the Mixer

Source: S02_firstImprov.wav (both excerpts) · guitar-side delay and pitch-shift pedals engaged ahead of the no-input mixer

These two excerpts document the guitar's own pre-mixer effects chain at work. With delay and pitch shifting engaged together, regularly spaced echo repeats show up as a comb of vertical bands across the whole spectrum. With pitch shifting alone, roughly five-second waves of rising and falling energy replace the discrete echo repeats, and none of the four bands shows a net gain or loss by the end (largest drift −1.7dB).

Interactions with the NIM

00 — Final Improvisation as Synthesis

Source: S02_fourthImprov.wav (23:11) · four fresh case studies from the session's longest take

This take confirms what the technique-by-technique studies above suggest separately: the system behaves again like a threshold system that answers to the shape and rhythm of re-excitation. Across these selected moments, the loop moves through nearly every regime documented throughout the session: a quiet reordering of which band leads, a multi-cycle breathing pattern that never resolves, a restless texture settling into a held tone, and a full collapse into silence at the very end. Moreover this take shows how if left alone for long enough, the loop's own accumulated state strongly influences which side of that threshold it lands on next, adapting the morph in relation with the next excitation given by instrument.

Interactions with the NIM

Setup

Amplification & No-Input Mixer Configuration

The flute is captured two ways at once: a contact mic on the instrument body and a DPA mic.

Mackie 1202-VLZ PRO

Channels 1 and 2 are self-patched through their own insert tip-send back into their own return, creating a feedback loop built directly into the Mackie's own patch bay.

Gain & EQ

Gain range depends on input type: 0dB to +60dB using XLR, using Line there is a 15dB attenuation up to 45dB at the knob fully up; the stereo channel runs from off +10dBV to +4dBU (no knobs).

Control techniques

Instrument techniques

Analysis method

Every recording below was processed the same way: onset detection (spectral-flux with backtracking), spectral-feature extraction (centroid, bandwidth, flatness, RMS) on a 2048-sample STFT, segmentation with clustering into acoustic classes, spectral-peak tracking to find recurring "attractor" frequencies, and isolated-onset case studies to see how different acoustic sound morphologies interact with the feedback loop.

General Overview & Synthesis

The flute and the no-input mixer

The session was structured in two stages: first, systematic, gesture-by-gesture studies that isolated one technique at a time; then, an improvisation that combined everything into a single continuous short performance. Total: 36.0 minutes, 8,267 detected onsets, seven recordings, nine techniques.

1. Different context, different outcome

All breath-and-embouchure techniques (constant tone, air sound, and roar recordings) show tracking into different frequency bands. Yet, in the single gesture session, this locking doesn't excite self-sustained regenerative buildup of the mixer: the feedback release dies at roughly the rate the air is withdrawn. Differently, the short, discrete, breath-free (or nearly so) gestures like pizzicato and key clicks show the opposite: examples of single brief attack triggering multi-second regenerative buildup, frequency-selective sustain, or sudden spontaneous collapse. In the longer, different session of the full improvisation, though, the picture is different: the constant-tone-like region shows a full regenerative buildup. The same technique that showed no such behaviour at all in its own dedicated recording shows a late, spontaneous high-frequency surge with nothing acoustically triggering it. Same technique, different context, different outcome.

2. A typology of feedback behaviours

Across both the isolated-decay case studies and the close onset-level pass documented below, the same handful of qualitatively distinct behaviours keep recurring:

No technique produced only one of these; every recording with more than one or two case studies shows at least two different behaviours.

3. How locked frequencies behave

The close-listening pass across all six technique recordings turns up at least four distinct patterns of the mixer locking onto a frequency.

A single, near-total hold: Two windows show one lock dominating essentially the entire span with nothing else registering: roar's core buzz (~159Hz) and one constant-tone window (~358Hz).

Two locks trading dominance: In the key-clicks passage and the roar passage, two different frequencies both register as "the strongest peak" at different moments within the same window, neither cleanly replacing the other.

A sequence of locks, one after another: Key clicks' longest excerpt moves through five distinct locks in order; whistle tones' clearest excerpt shows a three-step descent from roughly C9 through A8 to a settled E8 (note precision is rounded to the closest one). The loop finds, holds, and loses a resonance repeatedly across one continuous passage.

Simultaneity: two locks at once: In two singing+playing excerpts, two different locks are present over almost entirely overlapping time spans, not one after another, a direct evidence for that recording's two-independent-sources character. The same pattern turns up in whistle tones, away from any vocal involvement: two simultaneous locks (~1,950Hz and ~5,134Hz, a non-harmonic relationship ratio of 2.63) holding together for nearly nine seconds.

4. How technique and material shape the interaction

Register and noisiness separate the instrumental techniques cleanly. Plotting every recording by brightness (centroid) and noisiness (flatness) puts them into clearly separated territory. Roar, pizzicato, and key clicks cluster in the low-centroid, extremely-low-flatness corner, dark and almost perfectly tonal, despite being acoustically very different gestures. Constant tone/vibrato/bend/air and singing+playing sit in the middle. Whistle tones stands alone at the far bright end (6,041Hz).

The physical shape of an attack or release within the feedback loop is set by instrumental technique. Breath-and-embouchure techniques attack and release gradually, 2 to 6 seconds each way, scaling with how much of the gesture has to be physically established or released. Percussive/mechanical techniques (pizzicato, key clicks) attack in well under a tenth of a second. Whistle tones don't have a consistent shape (sometimes a slow, flickering 4-second establishment, sometimes a sub-second snap) this is directly connected with how unstable to control acoustically the technique is. In this scenario the mixer's reactions tend to mirror the boundaries of the physical gesture to then evolve by locking in/blending or morph independently.

5. Overall picture

A characteristic of these interactions between a wind instrument and the mixer's feedback underlies how the physical mechanism of the Flute's techniques can reliably determine the shape of its edges (how an attack or release begins and ends). What happens inside that edge, whether an attack triggers multi-second regeneration, locks onto a new frequency, redistributes its energy upward, or simply decays away, depends on the state the loop was already in when the attack landed, at least as much as on the gesture itself. The instrument's role is here closer to deciding which edge of a state shapes the loop excitation instead of determining what's inside it: the physical gesture governs the boundary and the feedback loop's own accumulated state governs everything the boundary contains. Again, this kind of setup tends to behave more as a threshold system.

01 — Constant Tone, Vibrato, Bend, Air

Source: S03_ConstantTone_Vib_Bend_Air.wav · 6:41.28 · four techniques in sequence: constant tone, vibrato, pitch bend, air sound

The recording has been divided into four blocks, each matching one of the four described techniques almost exactly in both duration and character: Block A is a simple clean tone, Block B is vibrato, Block C is pitch bend, and Block D is air sound.

Pitch content: a flute's pitch is set directly by the player, so the dominant frequencies here are best seen as which notes the player chose to play, not a property of the mixer. Two pitches dominate: 448Hz (roughly A4) and 525Hz (roughly C5).

Interactions

01 — Lock at the tail

1′07″–1′20″ · Block A

The attractor here (~557Hz, roughly C#5) is present for almost the entire window, with a second lock (~358Hz, roughly F4) only appearing in the final two and a half seconds.

02 — Most stable interval, lowest onset rate

3′03″–3′15″ · Block B

A single lock (~358Hz, roughly F4) spans the entire fourteen seconds shown. The centroid moves by under 450Hz in either direction and settles back each time.

03 — Onset real modulation

4′48″–4′56″ · Block C

The attractor (~358Hz, roughly F4) holds for the first ten seconds, but then suddenly glides low.

04 — Pitch and distortion shift together

6′08″–6′11″ · Block D

Three locks inside three seconds: starts near C#5, the dominant lock shifts to A#5 within the first quarter-second and holds there, with E6 flickering in briefly around the two-second mark. The two onsets that open the excerpt (t≈368.1s, 368.3s) show centroid and flatness jumping together (+654Hz/+0.012 and +469Hz/+0.011), pitch and distortion changing at the same instant. A third onset (t≈368.9s) reverses both while RMS rises settling into a pure, louder new state.

02 — Pizzicato

Source: S03_pizzicato.wav · 2:43.70 · a percussive, tongued attack

Pitch content: peak-tracking shows a noticeably wider, even spread of pitches.

Interactions

01 — No single outcome from the attacks

0′05″–0′26″

Peak content spreads continuously across roughly 150–1,200Hz. Outcomes vary onset to onset: the strongest attack in the window is followed by centroid dropping 562Hz and RMS dropping 2.2dB (darkening and quieting) while an earlier, weaker attack (t≈6.3s) brightens and loudens (+748Hz, +3.4dB).

02 — More concentrated, more settled

0′43″–0′50″

Peak content concentrates more here, with the ~358Hz region being the most visited. Onset effects are small and consistent: centroid moves by well under 150Hz either way across the six strongest onsets, RMS barely changes.

03 — A measurable transition into the pattern

1′48″–2′00″

Opens with two brief, very high transient locks (around C9 and C#9, each present for about a second) before settling onto a persistent low lock for the rest of the window.

03 — Roar

Source: S03_rorar.wav · 1:46.02 · embouchure fully covered, tongue rolled, producing a deep growling "roar"

Almost entirely continuous. The sound is strikingly low. A markedly different, brighter episode interrupts this around 56–65s before settling back into the low, pure register for the second half.

Pitch content (full recording): two dominant regions: 99Hz (roughly G2), present almost continuously, and 296Hz (roughly D4, its third harmonic). A secondary lock around 79Hz is also present through the first half.

Interactions

01 — One measured interruption

0′09″–0′21″

A single lock (~159Hz, roughly D#3) dominates virtually the entire twelve seconds. The one clear break in the pattern is precisely dated by the onset data: the strongest attack in the window (t≈16.9s, strength 3.14, roughly 7.9s into the excerpt) is followed by centroid jumping +455Hz and RMS rising 5.5dB, before the texture returns to the same low lock.

02 — Redistribution

1′01″–1′06″

Median centroid here looks like a completely different register from the core roar texture, but the attractor is still the same ~159Hz lock, present through nearly the whole window. The strongest shift (t≈64.8s) sends centroid up while RMS drops nearly 8dB: energy moving up into the harmonics of the same low anchor, not new loud broadband content arriving on top of it.

03 — Two locks trading dominance

1′31″–1′37″

~358Hz (roughly F4) is the more persistent lock here, with ~159Hz (roughly D#3) present but weaker. Onset effects swing both ways: one onset (t≈92.5s) brightens and the others darken, consistent with the core buzzing.

04 — Key Clicks

Source: S03_KeyClicks.wav · 4:46.49 · a purely mechanical attack — finger/key contact against the flute body, no breath involved

Almost entirely continuous. 1,797 onsets, mean rate 6.27/s. Flatness is very low throughout the take, each click still carries a clear resonant pitch from the air column. Centroid swings as different key/fingering combinations are worked through.

Interactions

01 — Two locks trading dominance

0′18″–0′50″

A low lock (~159Hz) is present almost continuously; a higher one (~557Hz) is nearly as persistent from about two seconds in. The two trade which is momentarily stronger rather than one cleanly replacing the other. Individual onsets push centroid substantially in both directions.

02 — Three adjacent locks, unstable late

0′58″–1′08″

Not one static pitch: the ring moves through three adjacent high locks over the ten seconds (roughly around G8 early, through E8, settling toward C#8 by the end). It isn't perfectly stable.

03 — Uniform, with one dated exception

1′42″–2′00″

A single lock (~159Hz) dominates essentially the entire excerpt. One onset (t≈115.1s) briefly breaks the pattern before the texture returns to the same low, clean lock.

04 — Five locks, the largest jumps of the session

3′20″–3′57″

Five distinct locks emerge and recede in sequence: around roughly C#9 early, down through C9, settling on C#8 for the bulk of the passage, then dropping to C8 late, all riding over a low ~159Hz lock present throughout. Two onsets in particular (t≈209.3s, 211.5s) are followed by centroid jumps up.

05 — Singing + Playing

Source: S03_SingingPlaying.wav · 2:59.37 · singing and playing the flute simultaneously

Direct inspection of the spectrum at several points through the recording consistently turns up multiple simultaneous, non-harmonically-related partials (e.g. at t=20–22s: prominent energy at 193.8Hz, 290.7Hz, 393.0Hz, and 430.7Hz together, none of them is an integer multiple of another).

Interactions

For each excerpt below, the key evidence for "two sources" is whether the attractors found inside the window have overlapping or sequential time spans.

01 — Two locks, almost entirely overlapping

0′16″–0′23″

Two locks (~358Hz and ~557Hz) are present over almost entirely overlapping time ranges (0–5.5s and 0.3–9s) rather than one replacing the other; this evidences the presence of two simultaneous sources. Flatness climbs steadily across the six strongest onsets, consistent with the two sources' interaction building in complexity.

02 — A register drop

0′50″–1′00″

Again two overlapping locks (~159Hz and ~358Hz).

03 — Voices enter apart, end in fast collapse

1′55″–2′02″

The ~358Hz lock is present almost from the start; a second, ~159Hz lock only joins partway through (from about 4.8s in): two voices entering at different times. The excerpt ends with a collapse: the final two onsets (t≈120.9s, 121.2s) are followed by RMS dropping 7dB and then 16.6dB in quick succession (a fade out).

04 — A bright opening

2′38″–2′48″

Opens with a sharp, loud brightening (the two onsets right at the start (t≈158.15s, 158.23s) before settling into the ~358Hz lock that carries the rest of the window.

06 — Whistle Tones

Source: S03_whislteTones.wav · 3:00.36 · a very soft, delicate embouchure technique that favours high overtones over the fundamental

The spectrogram shows a texture markedly more broken-up and discontinuous.

Interactions

01 — A measurable descending settle

0′30″–0′43″

Starts around roughly C9, drops through A8, and settles on E8 from about six seconds onward, holding there for the rest of the window.

02 — Two simultaneous locks, most settled

1′14″–1′23″

Not one stable pitch but two simultaneous locks (~1,950Hz and ~5,134Hz), both persisting for nearly the entire nine seconds. The ratio between them isn't a clean harmonic relationship, so these could be seen as the technique triggering an independent resonance rather than reinforcing the overtones over the fundamental.

03 — Five locks across nearly four octaves

1′35″–1′45″

Five different locks appear at different points, hopping across nearly four octaves. Adjacent onsets frequently push in opposite directions.

04 — Settling downward instead of up

2′26″–2′36″

Opens with the same kind of wide register-hopping as the first excerpt, but here it doesn't resolve into a bright, settled tone. Instead, three consecutive onsets late in the window (t≈153.4s, 153.9s, 154.2s) in the same direction: a sustained descending line that brings the texture to settle downward.

00 — Final Improvisation as Synthesis

14:05.60 · 2,868 onsets · median centroid 4,867Hz · free improvisation combining all nine techniques explored individually in this session

Interactions

Four gap-truncated isolated onsets, chosen from across different candidate regions, and together they cover most of the behavioural repertoire documented individually across this session:

Setup

Amplification & No-Input Mixer Configuration

The clarinet is captured with a condenser mic into Channel 3.

Mackie 1202-VLZ PRO

Gain & EQ

Gain range depends on input type: 0dB to +60dB using XLR, using Line there is a 15dB attenuation up to 45dB at the knob fully up; the stereo channel runs from off +10dBV to +4dBU (no knobs).

Instrument techniques

Analysis method

Each technique instance below is a short excerpt (7–22 seconds) cut from one of the raw recordings. Every excerpt is processed the same way: a log-frequency spectrogram (2048-sample STFT, 512-sample hop, native 44.1kHz sample rate) and a 4-band energy trend (80–200Hz, 200–500Hz, 1–2kHz, 4–8kHz), to characterise whether the mixer's feedback loop builds, decays, or holds steady during and immediately after the gesture.

General Overview & Synthesis

The clarinet and the no-input mixer

The session recorded six takes. The first five are sources of the seventeen excerpts documented on this page; the sixth is a final improvisation.

1. The instrument fingerprint

The session's own overview already establishes this at the full-recording level — multiphonics is the darkest and most tonal recording overall, teeth on reed the brightest, while noisiness itself actually peaks with overblowing. The seventeen excerpts on this page show how the system's stability its directly influenced by the sound production of the clarinet. Unlike the flute, the reed makes tones and attacks more controllable and less airy, allowing the clarinet to excite the mixer and almost control it. This sums up the acoustic morphology of the instrument (odd harmonics / square wave-like), creating a signal capable of triggering the mixer as well as directing the feedback or letting it start to then interact with it.

2. How technique and material shape the interaction

Complicate fingerings swings from buildup to collapse: Multiphonic IV gains 26.5 to 27.9dB across all four bands, one of the largest buildups documented in the session, while Multiphonic III, cut from the same recording only 35 seconds earlier, collapses by 8 to 15dB. The same fingering produces two opposite results based on the control applied over the tone sustain.

Breath is not required for the loop to regenerate: the Key Clicks recording's without-air condition, still triggers spontaneous multi-second regeneration, climbing 9 to 12dB across all four bands. Whatever catches and re-amplifies these brief mechanical taps is affected entirely by the air introduced.

Displacing the embouchure changes where the instability hits: the Teeth on the reed technique introduces other than breath control, direct mechanical contact, producing a bright, densest technique, unstable at the attack and only settling once the tone is established, the opposite order from overblowing, whose instability runs through the whole excerpt with no settling at all.

3. Overall picture

What decides the outcome here is not the reed's register but how its stability is disrupted. The same fingering can push the loop into very different results, and a purely mechanical tap with no breath at all can trigger the same kind of regeneration a sustained tone can. As with the other sessions, this behaves like a threshold system, except here the threshold responds to the reed's instability rather than to the note being played. Furthermore, the acoustic nature of the instrument is capable of more interconnected interactions with the system.

01 — Clean Tone

Source: CleanTone_S04.wav · slow attack, beatings, and short attack

The slow-attack excerpt, as expected, establishes more gradually. The two beatings excerpts both carry a fast, regular oscillation on top of the sustained tone, where the frequency band traces a dense, continuous zigzag for almost the entire eight-second window. The two short-attack excerpts both show rapid alternation between distinct spectral states: several short, percussive-adjacent attacks in quick succession.

Interactions with the NIM

02 — Overblown

Source: Overblown_S04.wav · the reed overblown into an unstable upper register

All three excerpts show the same jumpy, unsettled character. Nothing here could seem like as the loop developing something independently so much as directly carrying through the acoustic unpredictability of an overblown reed. The first and second excerpts swing repeatedly between loud and quiet across every band with no stretch settling into a lock. The third shows a distinct rhythmic pulsing: short, regular silences interspersed with bursts of held tone, a rougher and more mechanical-sounding pattern than the other two excerpts' continuous jumpiness.

Interactions with the NIM

03 — Multiphonic

Source: multiphonics_S04.wav · a special fingering producing multiple simultaneous pitches

The four multiphonic excerpts cover a wide range in their own right. The first shows a slow, wandering shift between registers rather than a single settled fingering. The second sits inside the recording's bright stretch and carries a clear, dense oscillation in the 200–500Hz band for several seconds. The third shows a rougher, faster, more chaotic version of the same oscillation alongside a real collapse, with every band losing 8–15dB. The fourth, after an unsettled opening, shows every band gaining between 15 and 28dB, a full-spectrum buildup roughly a hundredfold in power.

Interactions with the NIM

04 — Teeth on the Reed

Source: TeethOnReed_S04.wav · teeth placed directly on the reed instead of the normal embouchure

Both excerpts carry the dense, bright, buzzy texture. The first excerpt, closer to the start of the take, shows a genuine buildup: every band gains 12–21dB across the window, the reed's altered contact catching and amplifying rather than settling. The second, a little further into the recording, is calmer by comparison, but still dense and jagged from moment to moment, but with a smaller net gain (2–4dB across most bands).

Interactions with the NIM

05 — Key Clicks

Source: KeyClicks_S04.wav · purely mechanical finger/key contact against the body, with and without breath

The first excerpt shows the without-air condition: rather than a clean decay after each tap, every one of the four bands climbs steadily by 9–12dB across the ten-second window. The second is the odd one out: the brightest, densest excerpt in the technique, and the only one where every band declines rather than builds. The third, with-air, stays bright and dense throughout and keeps building regardless: all four bands still gain 7–10dB across the six-second window, breath keeping the loop fed independently of whatever the clicks themselves are doing.

Interactions with the NIM

00 — Final Improvisation as Synthesis

Source: Final_Impro_S04.wav (9:13) · four fresh case studies from the session's own closing improvisation

This take confirms, within a single unbroken performance, what the technique-by-technique studies above suggest separately: the loop again behaves like a threshold system, one that answers less to which note is played than to how stable or disrupted the reed's own excitation is at that moment. Across these four selected moments, the loop moves through nearly the full range of behaviour documented across the session: a dense, erratic fluctuation with no settled trend, a dramatic synchronized collapse followed by just as sharp a recovery, a steady sustain that tips into a late full-band brightening, and a clean, repeated alternation between a quiet state and a loud one, all with no change of technique behind any of these transitions. As in the individual recordings, breath and mechanical contact remain the two levers that decide which side of that threshold the loop lands on next: confirmation that it is the reed's stability, not its register, that keeps steering the interaction even across one continuous, freely moving take.

Interactions with the NIM