audioeffects.fuzz

Fuzz - a biased germanium pair, with a Big Muff cascade as a second character.

The player’s text is the class docstring; the CLAIMS table in the class’s test file ties its claim sentences to tests.

audioeffects.fuzz.GERMANIUM_SCALE = 0.632120559

What full scale on GERMANIUM_CURVE stands for. The table is normalised to the curve’s own extreme so it fills int16 (audiocomponents#77) and this carries the scale back out in the shaper’s post_gain, so the class’s level does not move.

audioeffects.fuzz.HP_HZ = (14.0, 7.9, 31.0)

The three coupling caps S1 names, in Hz. Each is one real pole in the circuit, and a biquad holds two, so the germanium input section is two nodes rather than the three Butterworth sections this class first shipped - which were sixth order where the pedal is third, and cost a node for it.

audioeffects.fuzz.HP_PARK_HZ = 5.0

The pair (7.9, 14) and the pair (31, PARK) as HIGH_PASS biquads. Two real poles at f1 and f2 under a double zero at DC are exactly two cascaded first-order high-passes: f0 = sqrt(f1 f2), Q = f0 / (f1 + f2). Literals, not math.sqrt at import: a filter setting derived in Python float differs between a board and a desktop (audiocomponents#75), and a literal parses the same on both.

HP_PARK_HZ is the spare pole of the output section. It goes below the band, not above it: a high-pass’s second zero already sits at DC, so a parked pole above the corner would tilt the whole band at 6 dB/octave.

Where below the band matters, because this section is what carries the clipper’s offset out. With zero in, the germanium curve still puts a constant at the shaper’s output, and this section is the only thing that removes it. A step into a double-zero pair of poles a < b leaves a slow term of amplitude a/(b-a) decaying at 1/a: parked at 1 Hz that is 3.3 % of the offset decaying over 159 ms, and the output was still moving 0.80 s after a burst stopped. At 5 Hz it is 19 % decaying over 32 ms - louder at once, gone sooner - and it costs 0.068 dB at 40 Hz and 0.011 dB at 100 Hz, where the first-order 31 Hz pole above it already costs 2.04 dB and 0.44 dB.

audioeffects.fuzz.CHARGE_BLOCKS = 32

How many blocks _charge_output will pull before it gives up, and how long the zero sample it pulls through is. The germanium output pole is 12.45 Hz, so its charge is about 0.15 s: 32 blocks of 512 frames is 16384, which covers it at every rate. The shaper’s native play() takes no loop, so the sample is a block long and is re-played each pull.

audioeffects.fuzz.CASCADE_OUT_HP_HZ = 31.0

cascade’s output coupling capacitor, in the graph since the fourth fix round (audiocomponents#89). A Big Muff has one between the second clipper and the tone stack; this class had none at all, and the low arm of the tone stack is a LOW_PASS, which passes DC. With Bias off centre the class stood -16 383 to +16 382 LSB (-6.0 dBFS) on the output for ever, at eight of the nine stops of the Bias macro, at every rate. 31 Hz is the pedal’s own corner and the one already in front of the second clipper. Measured against the class without it, on the wet branch at Tone 0.5: 1.268 dB at 40 Hz, 0.295 at 60, 0.053 at 100, 0.006 at 200 and 0.000 at 482 Hz and above - the same shape hp_mid already imposes one stage earlier.

audioeffects.fuzz.LEAN_PATCH = 8

Patch 8, Fuzz - lean. The factor is fixed at construction - the node cannot retune its half-bands live - so the lean position is a second shaper at LEAN_OVERSAMPLE, built beside the shipped one and played by the tilt stage when this patch is selected. Only the routed shaper is ever pulled; the other one costs its 5696 B of RAM and no time at all.

audioeffects.fuzz.GERMANIUM_OVERSAMPLE = 8

The factor each character ships at, and the reason it is not one number.

germanium is one clipper: ×8 prices at 2.257 / 3.980 ms and the boards measured 2.379 / 4.096 — rt 1.81 on the P4 and 1.05 on the S3, which clears. cascade is two clippers behind a tone stack, and at ×8 the same boards read 4.782 / 8.580 ms — 90 % of a P4 block and 161 % of an S3 one, rt 1.00 and 0.56 (audiocomponents#74). It does not run. The docstring said so before this round and nothing enforced it: oversample is a constructor option a user reaches, create() defaulted to 8 for both characters, and a default that cannot make the deadline on a board this library ships for is not a default.

×2, and it costs floor — said out loud. Cascade’s alias floor, 48 kHz, −6 dBFS, wet branch, A4’s own rule, measured for the first time this round, at 1010 / 3700 Hz:

Fuzz default ×1 −18.315/−13.407 ×2 −27.372/−21.343

×4 −30.123/−24.946 ×8 −30.232/−26.595

Fuzz maximum ×1 −17.537/−13.283 ×2 −24.237/−20.397

×4 −28.491/−23.897 ×8 −29.847/−25.982

So ×2 gives up 2.9 / 5.3 dB at the default and 5.6 / 5.6 dB at maximum Fuzz against ×8. ×4 would give almost all of it back (within 0.1 / 1.6 at the default, 1.4 / 2.1 at maximum) and it does not run either: 3.250 / 5.971 ms by the palette, 61 % of a P4 block and 112 % of an S3 one. ×2 is the only factor that makes the deadline on both boards.

Oversampling buys cascade much less than it buys germanium in the first place — ×8 over ×1 is 12.3 / 12.7 dB here against germanium’s 20.9 / 18.7 — because the second clipper folds harmonics the first one already put above the output Nyquist, which no internal rate reaches. Cascade is outside A4’s span at every factor (A4 is a germanium row: nothing in its sweep passes character=, and at ×8 cascade reads −29.8 / −26.0 against A4’s surface bars of 35 / 28 dB). That is disclosed here, in the class docstring, in the catalogue row and in the pack, in the same words.

audioeffects.fuzz.MIXER_BUFFER_BYTES = 1024

The tone mixer’s buffer, per channel. Mixer._render_size is buffer_size // 2 // 4 * 4 BYTES, so the 1024 cascade shipped rendered 128 stereo frames and the tilt stage behind it pulled the three-voice mixer, both tone arms and the dry tap twice per 256-frame block — the palette’s block, and the unit every audiodsp node works in. 1024 * channel_count is one pull per block at either channel count, which is what Distortion already does (_pcm_mixer).

audioeffects.fuzz.shipped_oversample(character)[source]

The factor this character ships at: ×8 on germanium, ×2 on cascade. Chosen on cascade’s own alias floor — see CASCADE_OVERSAMPLE, which has the numbers — and not on the maximum the node offers.

class audioeffects.fuzz.Fuzz(source, *options, **keywords)[source]

Bases: Component

Fuzz Face germanium, or a Ram’s Head 1973 Big Muff as cascade.

Character. character picks the pedal when you build it: germanium (the default) or cascade. On germanium the transistor sits off centre, so the fuzz leans even: at the default the second harmonic of a -20 dBFS 1 kHz tone is no more than 30 dB under it, and the even harmonics together no more than 26 dB. Turned up to 48 dB of Fuzz and past, both halves go square and the lean is gone. There is no tone stack on germanium: the band stays flat within 1 dB, and 40 Hz within 3 dB of 1 kHz.

Cascade. cascade is two silicon pairs with a mid scoop between them. With Tone and Bias centred the scoop is more than 3 dB deep. From Fuzz 18 dB to 32 dB, Tone moves the scoop’s centre over more than half an octave; from 34 dB up the clipper fills the scoop in.

Fuzz and Bias. Fuzz is how hard the transistor is driven, and Bias moves it off centre. The Effect controls page lists what the knobs span and the patches. On cascade any Bias off centre is a gate on the level: it rails both pairs and the output falls away, while the scoop stays more than 3 dB deep. With Bias off centre the output settles near zero rather than on it, so the class declares no tail there.

Level, Load and Mix. Level, Load and Mix set the volume, the load and the blend. Level 0 is digital silence at every Mix above 0, and halving Level takes 6.02 dB off at every Mix. Load 1 takes 6.02 dB off at every Mix. At Mix 0 the output is the source itself, at its own level, and the latency is 0.

Output Tilt. Output Tilt tilts the balance of lows and highs. Nothing on the surface reaches the rail from 100 Hz to 3700 Hz, even at 0 dBFS in.

Aliasing. At 48 kHz and 44.1 kHz, at the default, the aliasing is at least 50 dB under a 1010 Hz tone and 35 dB under a 3700 Hz one. Nowhere else on the surface, Fuzz - lean aside, does it come closer than 35 dB and 28 dB. At 22.05 kHz that does not hold, because the fuzz’s own harmonics fold back at the output rate.

Latency and tail. germanium shapes at x8 and reports 4 samples of latency; cascade shapes at x2 and reports 2, and so does Fuzz - lean. With Bias centred the tail is 24576 samples.

Patches. Alone on an ESP32-S3 the germanium default was measured at 77 % of a block and the costliest patch at 78 %. Fuzz - lean shapes at x2 and was measured at 33 %, so play it when you stack effects.

It needs audioshaper, audiobiquad and audioroute, so a stock CircuitPython board can’t run it.

NAME = 'Fuzz'

The stable provider name audioeffects.create() discovers.

TIER = 'audiodsp'

STOCK or AUDIODSP, with REQUIRES naming the audiodsp-own modules an AUDIODSP class needs.

Type:

roadmap §3

CAPABILITIES = ()

The contract’s static reads. CAPABILITIES names "tempo_sync" if and only if the class reads self._transport().

TAIL_SAMPLES = 24576

Measured, not inherited, and re-measured at 40 Hz by the third fix round: 16 384 was a 1 kHz number and a low note outlives it - 16 627 frames at the constructor default, 18 750 at patch 5, and 100 Hz is over it at patches 1, 5 and 7 (audit 3 (p)1). This covers the worst of 40, 100, 220 and 1000 Hz at the default and at patches 1, 5 and 7, at all three rates, after a -6 dBFS burst. It holds with Bias centred; off centre the output settles near zero rather than on it, and tail_samples is None there (see the docstring).

After a reset, a rejoin or a control move a few LSB of float residue can remain in the output beside an untouched copy, past tail_samples. See “Detector and coupling state” in docs/reference/component-api.md.

MACRO_LABELS = ('Fuzz', 'Level', 'Tone', 'Bias', 'Load', 'Mix', 'Output Tilt')

Knob names, one per macro index; one mode per label; the private engineering spans behind them; and the patches on the 0-127 grid.

WALKS_TO_BYPASS = True

A class sets this when its graph, at Mix 0, walks to the source it routes around: its mixer levels to the dry alone, on ramped mixers (ramp_levels). _route_around then lets the port switch once the walk has landed rather than at once. Mix 0 is still the source, byte for byte, from then on.

property tail_samples

TAIL_SAMPLES with Bias centred or Mix at 0, and None anywhere else.

Off centre the shaper answers silence with a constant, the coupling poles behind it settle into a wander of a few LSB rather than onto zero, and nothing bounds when (or whether) it stops: on germanium it was still there 6 s after the input at every Bias stop but the three either side of centre, and on cascade it settled at some stops and not at others. A bound there would be a guess, and the contract’s None is the honest answer. Mix 0 is the dry alone, and the wander is not in it.