strudel-mixing

name: strudel-mixing description: The craft of making a mix sound pro and clear — frequency carving so each voice owns a band, low-end (kick vs. bass pocket, mono sub, sub vs. mid-bass), gain-staging and headroom, sidechain/ducking, panning and stereo width, depth via reverb/delay (front-to-back layers), and why amateur mixes sound muddy or flat. Use when the mix is muddy, things are fighting, gain-staging, EQ, low end, sidechain, it sounds flat, it sounds cluttered, too much reverb, nothing has space, or voices are masking each other.

See [[strudel-sample-library]] § "Mixing pitfalls that read as not playing" for gain floors, VCSL amplitude ratios, and hpf over-cutting — this skill does NOT repeat those. It covers the broader craft of WHY a mix sounds pro.

The core principle: every voice owns a band

A professional mix is a frequency map. Each element claims a region; nothing fights for the same turf. The mix sounds clear not because everything is turned up, but because nothing is in anyone else's way.

• Full-range voices are the #1 culprit. A pad running 40 Hz–15 kHz competes with kick, bass, melody, and air simultaneously. That is four fights at once.
• Subtractive EQ beats additive. Cut the frequencies a voice doesn't need before boosting the ones it does. A hi-hat has no reason to exist below 800 Hz — high-pass it. A pad doesn't need 40–80 Hz — high-pass it. Each HPF frees headroom the kick and bass need.
• The mud zone is 200–500 Hz. Every instrument has some energy here. It accumulates fast. The fix: cut non-bass instruments in this range rather than boosting your way out of it.

Low end: the kick–bass relationship

The kick and bass together form the rhythmic and tonal spine. They must complement, not stack:

• Give each its pocket. Kick fundamental typically 50–65 Hz; bass fundamental 80–120 Hz. Boost the kick slightly at its fundamental, cut the bass there. Reverse on the bass. Neither should dominate the same Hz.
• Tune them to each other. A kick tuned to A and a bass root of D# will fight harmonically regardless of EQ. Tune the kick to the track's root (or the bass root note) so the two lock as one object.
• Sub (20–80 Hz) vs. mid-bass (80–200 Hz). Sub is felt, not heard on small speakers. Mid-bass is what gives weight on phones/laptops. A good mix has both: sub for the room, mid-bass for translation. A sine-wave sub alone vanishes on laptops — use triangle or sawtooth with lpf ~600–800 to get harmonics into the mid-bass band.
• Keep kick transient distinct. The kick's click lives at 2–5 kHz — that's what punches through a mix. The sub thud alone doesn't cut; the high-mid transient does. Don't low-pass the kick so hard it loses its attack (see [[strudel-sample-library]]).
• Mono bass below ~150–200 Hz is law. Our ears can't localize bass; a panned sub just causes phase cancellation when summed to mono. Keep kick and bass .pan(.5) (center). Width lives in mids and highs.

Gain-staging and headroom

• Target individual voices at −18 to −12 dBFS average. Leave peaks no hotter than −6 dBFS per track before summing. Stacking 8 voices at −6 dBFS each clips the mix bus.
• The stack accumulates. Ten voices at .gain(0.5) adds up. Work the summed level — if the overall mix is clipping or squashing, the fix is gain on individual voices, not just pulling the master down.
• Loudness target for streaming: −14 LUFS integrated. Don't chase loudness in the mix; leave headroom for mastering. A mix peaking at −3 dBFS going in has nowhere to go.
• Velocity / gain variation is dynamics. A flat .gain(0.5) on every hit is metronomic. Use .gain(".8 .4 .6 .4") or gain(perlin.range(.3,.6)) for natural dynamics that preserve headroom by breathing.

Sidechain / ducking (the pump and the pocket)

Sidechain is not just a stylistic pump — it's a mixing tool that prevents kick and bass from arriving simultaneously at full volume:

• Strudel's method: duckorbit(N) on the kick triggers an envelope that ducks whatever is on .orbit(N). Pair with .duckattack(0.01) (fast, transparent) or .duckattack(0.2) (slow, audible pump). .duckdepth(1) = full duck.

// Kick ducks the bass on orbit 2
stack(
  s("bd!4").bank("RolandTR909").duckorbit(2).duckattack(0.01),
  note("C2 ~ Eb2 ~").s("sawtooth").lpf(400).orbit(2).gain(0.7)
)

• Audible pump (house/techno) → slow attack (~0.1–0.3s), deep depth → the bass visibly ducks on every kick. This IS the groove in much dance music.
• Transparent pocket (jazz, ambient) → fast attack (<10ms), shallow depth (~0.3–0.5) → just creates space, imperceptible as an effect.
• .compressor() in Strudel (syntax: compressor("threshold:ratio:knee:attack:release")) compresses a voice's own dynamics but does NOT do cross-signal sidechain — use duckorbit for kick→bass ducking.

Panning and stereo width

• The center is sacred: kick, bass, lead. These anchor the mix. Spreading them wide = phase issues and lost punch.
• Width lives in the mids and highs. Hats, shakers, pads, reverb returns, secondary percussion → spread these. They're what makes a mix wide.
• Avoid the "everything hard-panned" trap. Extreme L/R with nothing in the mid feels hollow. Pan in graduated steps: .pan(.3) / .pan(.7) creates width without emptying the center.
• Stereo width from detune, not pan. .add(note("0,.07")) or layering a voice slightly detuned fills width without phase risk. This is the Kiasmos/Rone pad-fattening technique — see [[strudel-pro-tips]].
• Check in mono. If a wide element collapses or disappears in mono, it has phase cancellation. Mono compatibility is non-negotiable for streaming/club.

Depth: front-to-back layering with reverb and delay

A flat mix has everything at the same perceived distance. A pro mix has a z-axis:

• Pre-delay is the trick professionals use. Adding ~15–20ms pre-delay before the reverb keeps the dry signal present while the tail floats behind it. Without pre-delay, reverb sits on top of the dry signal and muddies the attack.
• Reverb on the low end = mud. High-pass the reverb send at 200–300 Hz. Don't reverb the kick or bass — it blooms in the mud zone and destroys punch.
• Delay as width, not clutter. A short (1/8th or 1/16th) stereo delay panned L/R on a melody fills width and depth without adding frequency content. See [[strudel-effects]] for .delay() / .delaytime() / .delayfeedback() mechanics.

Strudel's EQ toolkit and its limits

Strudel has no per-band parametric EQ (no surgical 3 dB cut at 340 Hz). The workarounds:

• .lpf(N) — removes everything above N. Use to keep voices band-limited and out of the highs.
• .hpf(N) — removes everything below N. The most powerful mud-fighting tool. Apply to pads, melody, shakers — anything that doesn't need low-end.
• .bpf(N).bpq(Q) — bandpass; useful for telephone/lo-fi textures or targeting a vocal formant.
• Stacking lpf + hpf = a crude bandpass. .hpf(300).lpf(3000) limits a pad to the mid band only.
• Resonance (.lpq(), .hpq()) = emphasis at the cutoff. Useful for filter sweep character; dangerous if it pushes an already-crowded band higher.
• Full syntax for all filter/effect parameters → [[strudel-effects]].

The five things that make amateur mixes sound muddy

1. Full-range pads. A sustained synth pad with no HPF occupies the entire spectrum. It's not a pad — it's a wall. HPF at 200–400 Hz and it becomes a pad.
2. Low-mid accumulation. Kick + bass + piano + guitar + pad all have energy at 200–400 Hz. None of them need all of it. Cut each slightly in this zone.
3. Reverb without pre-delay (or on the bass). Reverb on bass = boom in the mud zone. Reverb without pre-delay = smear on the attack.
4. No gain variation across voices. Every voice at equal loudness = no hierarchy. The ear doesn't know what to listen to, so it hears noise.
5. Center-only mixing with no z-axis. Everything panned center and equally close = flat wall. Spread panning + reverb depth = three-dimensional mix.

Loudness & master level (Strudel does NOT auto-normalize)

Per-voice .gain() values ARE the master level — nothing fills the headroom for you. Conservative gains (a pad at 0.08, a mix peaking ~0.4) sound soft because the top half of the range sits unused. To make a track louder:

• Scale per-voice — that's the reliable lever. Multiply each voice's .gain() up. It multiplies linearly and preserves balance. Render and check peak < ~0.9 (1.0 = clip).
• A master .gain() on the outer stack(...) is NOT a clean trim — in this engine it behaved pathologically (clipping + harsh flatness at 1.3 and 2.0, identical; it blasts even near-silent voices like hats to the ceiling instead of scaling the summed bus). Don't reach for stack(...).gain(x) as a master fader. Scale voices, or use a real limiter (.compressor()).
• The loudest voice (usually the kick) caps the global boost. Pushing a kick past 1.0 gain distorts it. If a uniform scale sends the kick over, cap it (0.8) and let the other voices come up around it.
• "Harsh" flatness on a louder mix is usually the [[silence artifact]], not real distortion. Flatness is scale-invariant, so a uniform level change can't truly raise it — check peak first: if peak < 1.0 nothing's clipping, and the high flatness is the near-silent reverb-tail frames (amplified when louder loud-sections widen the dynamic gap). Fix by narrowing dynamics (lift the quiet-section floors), not by darkening. (tarn: per-voice ×1.6 read flatness 0.09 / dyn 1408× with peak 0.56 — lifting the intro/breakdown/outro floors snapped it back to 0.0001 / dyn 3.6, same loudness.)

Measure it: warm vs harsh vs dull (the ears)

Don't trust a mix you can't hear cleanly — measure it. Render the track headlessly (node tools/render-wav.mjs <id> → WAV) and read its spectral profile (tools/measure-wav.py), then compare to the reference card. Two numbers diagnose most problems:

• Spectral flatness = noise / harshness. Clean, tonal mixes sit ~0.001–0.02. Above ~0.05 the mix is noise-like — almost always too many bright voices stacked in one band (or open hi-hats). This is the single clearest harsh-vs-clean signal.
• Spectral centroid = bright vs dull. Aim for the artist's card zone. Too high → harsh/brittle; too low → dull/muffled.

Worked example — this project's own cranks, all measured:

The lesson: warmth ≠ dullness, and brightness ≠ harshness — they're different axes.

• Harsh = high flatness (noise). Fix by THINNING — fewer bright voices, carve them into separate bands — not by darkening. (tarn: pushing 808 hi-hats up for "presence" spiked flatness 0.0001 → 0.07, cobalt-harsh range. Hats are noise — they buy centroid and harshness together. Pulling them to near-silent texture restored 0.0001.)
• Dull = low centroid from harmonically-poor waveforms (bare triangle/sine). Fix by reaching for rich REAL instruments — gm_string_ensemble, piano, gm_cello, gm_koto, gm_flute — which carry presence in the low-mids while staying warm. Not by adding raw-synth brightness (→ harsh) or a noisy sparkle layer (→ flatness spike; a high triangle arp once pushed grotto from 0.0001 to 0.18).
  • A FILTERED saw/square is also a valid tonal centroid-lifter (the "raw-synth brightness → harsh" warning is about unfiltered / high-cutoff synths). tarn's dull triangle hook (centroid 318) → sawtooth under lpf(960) lifted centroid to 465 with flatness still 0.0001: a low-passed saw adds warm mid-harmonics (centroid up) with no high-freq noise (flatness flat). Put the brightness on a LOUD tonal voice (the bass hook/pad), since loud dark voices are what pin the centroid down.
• Gotcha when measuring a whole dynamic ARC (not a steady loop): near-silent + heavily-reverbed sections inflate the average flatness — diffuse reverb tails are broadband, so silence-drenched-in-reverb reads as "noise." If flatness is high but no single section sounds noisy, suspect reverb-wash-in-silence (tarn hit 0.09 at 477× dynamics this way). Fix: keep quiet sections quiet but DRY/tonal (cut room+delay-feedback in the dip), not washed.
  • The right dyn target is GENRE-dependent — don't chase a big arc on level-steady music. Ambient/cinematic wants high dyn (true-silence intros/breakdowns); but a four-on-the-floor dance track is meant to sit at a steady level, so a low dyn (~5×) is the correct end-state — its energy arc is carried by layers dropping in/out (drums out in the breakdown, half-time outro), not by RMS swings the meter sees. Hollowing the breakdown to win a bigger dyn number just re-spikes flatness (the silence/wash artifact). neon (STRFKR 4×4): a present breakdown held flatness 0.02 at dyn 4.9; lowering the breakdown to chase dynamics shot it to dyn 509× AND flatness 0.075 (reproducible across re-renders). Chose the present breakdown — clean + genre-right.
• The pro recipe (gyre/glade, both loved): real warm instruments + ruthless tonal discipline (no noise layers) + every voice carved into its own band + filter movement + a big dynamic arc.

Related skills

Compositional structure → [[strudel-arrangement]]; the dynamic arc across sections → [[strudel-transitions]]; effects syntax → [[strudel-effects]]; sample gain floors and VCSL levels → [[strudel-sample-library]]; pro technique shortcuts → [[strudel-pro-tips]]; sound design (synth timbres to layer) → [[strudel-sound-design]]; genre-specific treatment → [[strudel-genres]].

Sources: Production Music Live — EQ Kick & Bass · LANDR — EQ Kick and Bass · Sonarworks — Eliminate Competition Between Kick and Bass · Sonarworks — Sidechain Compression · iZotope — What is Sidechain Compression · LANDR — Headroom in Audio · Flotown Mastering — Center That Sub · Dead Set Live — Why Your Mix Is Muddy · DemoMentor — 13 Mixing Mistakes · Strudel — Audio Effects · Strudel forum — Sidechain in Strudel · Strudel PR #1470 — duck/duckorbit