What a partial actually is
Every periodic sound can be written as a sum of sine waves. The lowest is the fundamental, and every other discrete component is a partial. For a sawtooth at 220 Hz the partials sit at 440, 660, 880 Hz and upward, with amplitudes falling at roughly one over n. That single fact explains why a saw sounds bright and a sine sounds like a test signal, and it is the entire description of what an additive synthesiser does.
So the first question is not which synth to open but which spectrum you want. A hollow flute-like tone is a small set of partials at roughly 1, 3 and 5. A clarinet is close to a square wave and therefore odd partials only. A trumpet in a sustained note has strong 1, 2, 3, 5 and 8. These are measurable: record a solo instrument in a quiet space, take one stable FFT frame at a held pitch, and read the amplitudes straight off the display.
| Source | Dominant partials relative to the fundamental |
|---|---|
| Sine / test tone | 1 only |
| Flute | 1, 2, 3, most energy below 8 |
| Clarinet | 1, 3, 5, 7, 9 |
| Sustained trumpet | 1, 2, 3, 5, 8 |
| Ideal sawtooth | All, amplitudes at 1/n |
One over n is the closest a synthesis technique comes to a physical law, and it is easy to misapply. Real instruments deviate from it and the deviations are the character. Treat it as a starting point, then move individual partials a few dB by ear or by measurement and listen to what happens.
The maths you actually need
No degree required. Each partial is a sine oscillator whose frequency is n times the fundamental, sitting behind its own gain stage, and the sum goes to the output. The moment you modulate the amplitude or frequency of any of those oscillators, the static spectrum stops being a list and starts being an instrument, and additive synthesis begins to resemble the other methods.
A violin section playing one note with independent vibrato produces a spectrum whose partials smear sideways in frequency. That smearing is the cue your ear uses for many players rather than one machine, and it is free to add: put three or four cents of slow random detune on each partial and the same patch reads as a section.
- Set the fundamental first, then build partials upward. Skipping 2 and 3 leaves a gap the ear reads as a hole.
- Sweep the harmonic number over time rather than switching oscillators on and off. Hard gates click.
- Give upper partials a slightly later and shorter envelope. Real bodies shed their top end faster than their bottom.
Why it sounds thin, and what fixes it
The honest trade-off is that additive is precise and fragile. A sawtooth has a fixed spectrum too, but it has a filter, a distortion path and decades of people listening to it, so it sits convincingly in a mix. Thirty-two oscillators with no movement sit in front of the listener instead. Two changes close most of that gap: a few cents of independent detune on every partial, and a short envelope per partial.
The second fix is subtractive in disguise. Sum the partials, run the result into a resonant low-pass, and sweep the cutoff, and you have rebuilt a familiar synth voice entirely out of addition. That is not a criticism of the method. It is how the two approaches meet, and it is also how most virtual-analogue and wavetable instruments are actually constructed.
Level is the other surprise. A spectrum made of many discrete components cancels against itself in the waveform, so a patch that sounds loud can measure surprisingly low on a peak meter. Expect it, and gain stage the output rather than reaching for the channel fader later.
When breaking the pattern is right
Perfect spectra can be a mixing problem. Because every partial is a narrow event with a high crest factor, the result can read as thin and brittle next to anything with noise and saturation in it. Some of the best-sounding patches come from deliberately breaking the pattern: drop one partial by 6 dB, add a detuned octave below the fundamental, or let two partials collide at the same frequency and beat.
Measurement closes the loop. Render a bar and look at the spectrum. If the harmonic series is where you placed it, the tools agree with you. If the sound is not landing, the spectrum will say so before your mood does, and there is no guesswork left in the argument.
The bottom line
Additive is the only synthesis method where the spectrum is fully yours to specify, which makes it precise and unusually easy to get wrong. Used as a starting point and then broken deliberately, it is the fastest route to a spectrum you can measure rather than guess at.
Hear every partial in context
432Hz MASTER moves the whole production by one ratio, so the partial relationships inside any additive patch stay intact while you judge them.
Frequently asked questions
What is the difference between a partial and a harmonic?
A harmonic is a partial whose frequency is an integer multiple of the fundamental, so partial 3 is exactly 3 times the fundamental. A partial is any discrete component, and in a real instrument the upper ones are usually not exact integers of the fundamental.
Why does additive synthesis sound thin?
Because a static spectrum has no movement inside it. Real instruments detune, decay upper partials faster than lower ones, and sit in a room. Adding a few cents of independent detune per partial and staggering the envelopes is usually enough to close the gap.
Is a sawtooth really just 1 over n partials?
Close enough to be a useful starting point and not accurate enough to be a law. The upper partials of a real saw fall off faster than 1/n, which is why additive saws can sound slightly harsh until you tilt the top of the series down.
How many oscillators can I realistically run per voice?
On current hardware, 32 to 64 sine oscillators per voice is comfortable. Beyond about 64 you are paying for oscillators that contribute nothing audible below the noise floor of the mix.