This skill's stance: timbre is an arrangement decision, not a post-production matter.
Choose the wrong timbre, and all the subsequent orchestration, layering, and mixing is just damage control. And in most people's hands, the act of "choosing a timbre" is actually "flipping through presets until one doesn't sound bad"—that is precisely the most reliable source of AI flavor, because generative models are doing the same thing: taking the most typical timbre in that category.
The core method of this skill comes from Takao Nozaki:first state the concept clearly, then touch the parameters.
Which sections to read by task
| Symptom／Task | Read |
|---|---|
| Don't know where to start designing a timbre | §1 Four-question checklist → §3 Mapping |
| Flipped through presets for ages and none are right | §2 Return to zero point |
| Timbre "sounds like a preset," no personality | §2 ＋ §4 |
| The timbre is right, but it's wrong once placed in the arrangement | §3's ② Range |
| The bass disappears on phones/laptops | §6 Difference tones |
| The Pad is thick but has no texture | §4 Four steps of sculpting |
| Fill in the roster timbre descriptions of ARR-SPEC | §7 |
Boundaries: what this skill does not cover
| Not covered here | Covered where |
|---|---|
| Selection of real instruments and hybrid timbres | mc-orchestration |
| Which frequency band this timbre occupies, and who it clashes with | mc-texture-layering |
| EQ frequencies, compression ratios, bus processing | mc-mix-intent(intent)／outbound mixing library (practice) |
| Film/TV sound effects, Foley, ambience | Outbound → Sound Design and Mixing Library (same name, different thing, don't mix them up） |
| Timbre recipes for specific styles (how to dial in a Detroit techno bass) | reference.md 's recipe table |
| Sample selection and slicing | mc-style-hiphop §6.3–6.5 and its reference §1–2 (material selection, slicing, loop building); sample library product selection outbound |
★ Naming reminder: this library's
mc-sound-designrefers tosynthesis and design of musical timbres; "sound design" in the film/TV sense (sound effects, Foley, space) isanother library。
1. Must fill in before starting: the four-question checklist
Before starting the design, first determine the timbre'srole personality, and answer these 4 items (p0091):
| # | Question |
|---|---|
| ① | Does it belong tothe percussion instrument type(indefinite pitch) ormusical tones(has pitch, can play melodies or chords)? |
| ② | Roughly whatpitch rangeis it? |
| ③ | Which curve does this timbre'samplitude envelopepresent? |
| ④ | Bright (many overtones) or dull (few overtones)？ |
The author's requirement is to build a habit:"When engaging in timbre design, first build the habit of comparing your concept against ①–④" (p0091)。
Why these four questions are enough: They happen to correspond one-to-one with a module of the synthesizer (§3), so once you have answered the four questions, how to set the parameters is already more than half decided—the rest is taste。
Self-check action: If you cannot answer even one of the four questions, do not touch the knobs. When you cannot answer, what you are doing is not design, it is trial and error.
2. Return to zero point: start from the "unadorned sound"
Problem: A synthesizer, unlike an electric guitar or a piano, does not have a simple, clear "original sound." The moment you turn it on, you are already inside someone's preset,and every decision you make is made relative to someone else's decisions。
Method: Create an initial setting for an "unadorned sound" (p0088):
| Module | Setting |
|---|---|
| Oscillator | Use only one sawtooth wave |
| Filter | Turn CUTOFF fully clockwise, RESONANCE 0, FILTER EG 0(fully open) |
| AMP EG | ATTACK 0, DECAY 0, SUSTAIN fully open, RELEASE 0(organ-like envelope) |
| Modulation such as LFO | Do not use at all |
| Effects | Off |
"Only by setting it to this kind of 'state of the raw waveform sounding forth' will it form an 'unadorned sound' like a guitar or piano" (p0088)。
Suggestion: Save it as "Initial State No. 1,"and start every sound design from here (p0089)。
The mechanism by which this fights an AI flavor: A preset is someone else's average judgment about a category. Starting from a preset, you can only fine-tune around that average. Starting from zero point, your sound grows out of yourself.
When it fails: When time is extremely tight, or when you need a clearly defined classic sound (such as a TB-303 acid bass), starting from the closest preset and modifying it is reasonable (the method in p0090 ②)—butyou must be able to say which four places you changed, and if you cannot say, then you did not change it.
3. Four questions → parameter mapping
| Concept | Parameter | How to set |
|---|---|---|
| ① Percussive or tonal | Oscillator waveform | Percussive type → NOISE; tonal type → SAW or SQUARE (p0092) |
| ② Range | TUNE / TRANSPOSE | See ★ below |
| ③ Envelope curve | AMP EG | See the three groups below |
| ④ Bright or dull | FILTER | First set −24 dB/oct low-pass, then use CUTOFF to adjust (p0092) |
★ ② Range is the item most easily overlooked and with the greatest cost
"No matter how hard you try to approach the ideal sound, as long as the range is different, the sound will absolutely never be as you imagined" (p0092)。
This explains a common phenomenon:a sound sounds right on its own, but wrong once placed into an arrangement.
Most of the time it is not a problem with the sound, but that it is in the wrong register.
—So this connects directly to mc-texture-layering the "staggering registers" of §4.2.
③ The three groups of envelope decisions (p0092)
| What you want | How to set | |
|---|---|---|
| Attack | Fast and crisp / Gradual | Shorten the ATTACK / Lengthen it |
| Sustain | Fades away naturally / Continues to extend | SUSTAIN=0 and adjust DECAY / Use SUSTAIN to set the volume |
| Release | Disappears immediately / Leaves a lingering tail | RELEASE=0 / Lengthen the RELEASE |
General principle: "The most important thing is to start by working on the『already established parts』" (p0093). Among the four questions, the one you are most certain about is the first module you should touch.
Technique: If the phrase is already written,design the timbre while playing it back repeatedly; the effect is quite good (p0093). ——Timbre is judged in context, not by listening in solo.
3.5 ★ The attack is the timbre's ID card (Howard & Angus)
The four-question checklist asks about spectrum and envelope, butrecognizability is almost entirely in the first few tens of milliseconds: once the attack and decay are removed, a violin and a trumpet are hard to distinguish (p0222–0223). When a synthesized timbre sounds "fake," it is usually not the spectrum that is wrong, but the cues that should be in the attack are missing. The book names six (p0223): bow-string friction, the consonant-like attack of brass, the breath noise of wind instruments, the slap of a reed's attack, the strike of a hammer, and the fall-back of the string.
Three points when making timbres：
- To imitate a real instrument, first ask what noise or non-harmonic components are in its attack, then ask about the steady-state spectrum. A sample sounds "fake" mostly because the looped section is taken from the steady state and the attack has been flattened (p0217)
- The attack is not colored by the room; the first reflection usually arrives only after the attack has ended (p0217).Reverb cannot save an attack that has been flattened
- Attack lengths differ greatly between families: brass leads with the fundamental and is extremely short; the flute has about 70 ms of breath noise before the fundamental; the violin's attack is about 160 ms, and vibrato does not appear until after 250 ms (p0220–0221). To make a timbre "for fast staccato," the attack must be short; to make a timbre "for long notes," a longer attack actually sounds more convincing
The comparison table of attacks for each family is in mc-texture-layering reference §5a.2。
3.6 ★ Emphasis of sonic dynamics: give a fast attack to a stack with a slow attack (Yang Liqing)
"Sonic dynamics" refers to the temporal transition characteristics of a note's attack, steady state, and decay.Layering different timbres to strengthen, exaggerate, or even transform the original sonic dynamics of a given factor, this is called emphasis of sonic dynamics (Yang Liqing, Volume 2, p0287).
The most common use:a pad or sustained strings has no attack transient, so layer something with a fast attack (pluck, piano, pizzicato, mallet) to give it an attack transient, and perceptually the pad itself gains edges, though the attack transient actually comes from the other instrument. Conversely, an attack that is too hard can be layered with a slow attack to "soften" it.
This is one of the most common mechanisms behind "why two layers sound better together than each does alone." When writing it into the spec sheet, treat it as an explicit decision:
roster[].note write clearly "which layer is supplying the attack transient for whom."
In the same system there is another point related to timbre:the high point of psychological dynamics is not necessarily the high point of physical loudness(p0283–0284). For a melody's top note or structural note, even if the dynamics are moving downward, another instrument's timbre is often layered on to emphasize it. Berlioz layering a horn on the top note at ppp is exactly this idea. When making timbre layers, reserving an instrument's timbre that can be layered onto "the point that needs to be heard" is more effective than pushing the overall dynamics.
4. Sculpting the Model: What Subtractive Synthesis Is Doing
The author likens sound design tosculpting（p0037）：
| Sculpting | synthesizer |
|---|---|
| ① Choosing the material (wood, stone) | Choose the waveform |
| ② Using a composite material | Use multiple sets of oscillators |
| ③ From a massive prototype,carve awaythe blocks you want to discard | Cut off unwanted overtones with a filter |
| ④ The parts you need,patch and reinforce | RESONANCE highlight a specified frequency band, or use FM / OSC SYNC / RING MOD to create new overtones |
And EG and LFO create changes over time, akin to stop-motion clay animation (p0037).
The use of this model: it removes "adding effects" from the process. All four means in ④ are inside the synthesizer—get the sound right at the source first, then consider effects。
5. Filter Selection
| Type | What it does (p0037–0038) |
|---|---|
| Low-pass LPF | Cuts off starting from the top of the sound (high-frequency overtones) |
| High-pass HPF | Cuts off starting from the lowest foundation (low frequencies) |
| Band-pass BPF | Keeps only the wanted part, cutting both above and below |
| Notch | Cuts out only the specified part, keeping both above and below |
| Comb | Like a comb brushed through, leaving a striped result |
More than ninety percent of synthesizer sound design uses low-pass—because synthesizer preset waveforms usually contain more overtones than ordinary instruments, so high-frequency overtones need to be cut away (p0038).
So: if you are unsure which to use, use low-pass. When using Notch or Comb, you should be able to say why.
6. ★ Difference Tones: Why a Pure Sine Bass Disappears
This is apsychoacoustichard constraint, not a matter of taste.
Difference tone: even if the fundamental is inaudible, the brain infers the fundamental from the overtones. For example, hearing 300 Hz and 400 Hz at the same time, the brain automatically fills in the greatest common divisor 100 Hz, making one mistakenly think they heard a 100 Hz sound (p0127).
But—"A sine wave contains no overtones that can serve as a basis for judgment, so naturally no auditory error occurs" (p0127)。
Consequence: a pure sine bass sounds good on monitor speakers,but simply does not exist on phones, laptops, or Bluetooth speakers. Because those devices cannot reproduce the fundamental, and the sine wave leaves no clue for the brain to infer the fundamental.
Two paths (p0127)：
- Add overtones, creating a sound that can produce an auditory error ← almost always choose this one
- Assuming an environment capable of playing super-low bass, such as a large dance hall/concert hall,give up on ordinary listening/viewing environments
Two ways to add overtones
First: transposed layering (p0127) Copy the same performance to another track,TRANSPOSE it up an octave, or up an octave plus a perfect fifth, mix in slightly.
"Adding multiple overtones is also quite effective, but adding too many will lose the 'texture of a sine wave,' so aim for a level where they're audible but not quite audible" (p0127)。
Second: Distortion(Recommended for beginners, the effect is more natural) (p0127) Insert a distortion effect to create a subtle effect.
"If you can tell it's distorted, you've overdone it" (p0127)。
Routing method: send the sine bass via Bus to an AUX track, add distortion on the AUX and send it back to mix; if the amount of distortion is too small,first low cut on the AUX, so it won't interfere with the original sound (p0128)。
Self-check action: any arrangement that uses a sine wave/very dark timbre as the low-frequency main body must be verified once on a phone speaker. This is the only check in this skill that requires actually listening on a device.
7. Correspondence with ARR-SPEC / each backend
The timbre falls under ARR-SPEC's roster[].instrument(abstract choice),roster[].timbre(for text backends)
and roster[].synth_params(for parameter backends).
The same description must be consumable by all three types of backends, so how it's written matters:
| Backend | How it consumes | Writing requirements |
|---|---|---|
| Suno / MiniMax (text prompts) | Only understands natural language | Useaudible descriptors: material, era, reference instruments. Do not write CUTOFF values |
| YuE2 (text + conditions) | Same as above, can carry section-level differences | Same as above, but can differ section by section |
| MIDI / MusicXML → DAW | Understands parameters | roster[].synth_params Write the answers to the four questions and the key knobs |
Writing example(two projections of the same timbre):
- Text backend:
warm analog saw bass, slightly detuned, short attack, filter opens over 2 bars - Parameter backend:
{osc: saw×2 detune 8c, filter: LP24 cutoff 320Hz, FEG amount +40 decay 900ms, amp: A0 D200 S0.7 R120}
These two must share the same source—generated from the answers to the four questions, not written separately. If they don't share the same source, you'll get "the text backend and the MIDI backend sound like two different songs."
8. Must-pass after writing
- Every designed timbre has answeredthe four questions, and the answers are written down
- Starting fromzero; if modified from a preset, can say which parts were changed
- Rangewas decided first, not adjusted at the end
- Where a non-low-pass filter is used, can say why
- ④'s overtone enhancement is preferentially doneinside the sound source(FM/SYNC/RING/RESO), not by first attaching an effect
- If the low-frequency main body is a sine wave/very dark timbre,overtones were added, andverified on a phone speaker
- The roster's timbre descriptions provide botha text version and a parameter version, and the two share the same source
- The timbre wastested in the phrase context, not decided by solo audition
Appendix: Sources
-
Takahiro Nozaki, "Illustrated Synthesizer Music Creation Method"— §1–§6 all main sections (24 Theorems of timbre design, 25 Starting from the initial setup, 09 Filter ①, 36 Sine wave bass/difference tone). Page numbers refer to pages in that book.
-
Specific timbre recipes for each style (electro funk gritty bass, Detroit techno, IDM, Electro House, hip-hop fat bass, R&B ballad, drum and bass, etc., 10+ kinds) are compiled in reference.md。
-
Kazuhiro Takeuchi, EDM Production, EQ/compression chapternot in this skill—belongs to
mc-mix-intentand the outbound mixing library. -
For the acoustic principles of difference tones, see also Howard & Angus, Acoustics and Psychoacoustics of Music(missing fundamental).