What Sieve Does

Sieve is a real-time digital signal processing algorithm that isolates natural harmonics from audio signals using spectral masking. It does not generate new pitched content. It extracts harmonics already present in the signal.

Every pitch and harmony effect on the market today (octave pedals, pitch shifters, harmony generators) creates new signal content by transposing or synthesizing new pitches. None of them analyze and selectively extract the existing harmonic partials from the source audio. Sieve does.

The result is a new class of audio effect with a sound character that has never existed in a commercial product: the natural bell-like quality of acoustic string harmonics, available electronically, in real time, on any instrument, controlled by an expression pedal.

The Physics Behind It

When a guitarist lightly touches a vibrating string at a node point (a position where a specific standing wave mode has zero displacement), only the harmonic modes that have a natural node at that position can survive. All other modes are physically damped. This is how natural harmonics work on any string instrument.

The position of the touch determines which harmonics survive. Touching at the midpoint (12th fret, L/2) kills all odd harmonics and leaves the even ones: 2nd, 4th, 6th... Touching at one-third of the string length (7th fret, L/3) leaves only multiples of 3: 3rd, 6th, 9th...

Figure 6. Standing wave modes of a vibrating string. Touching at position L/n damps all modes that do not have a node at that point. Only harmonics whose numbers are multiples of n survive. Sieve simulates this electronically.

Between pedal positions, the algorithm interpolates continuously between adjacent node divisors. The transition between nodes produces a smooth spectral morphing effect as harmonics gradually fade in and out.

How It Works

The algorithm consists of five stages applied to each audio frame in real time at 44.1 kHz. The total pipeline latency is approximately 46ms (4096-sample window at 44.1 kHz), which is within the acceptable range for live performance effects.

Figure 1. System block diagram. Audio flows from input through pitch detection, STFT analysis, spectral mask construction, mask application, and inverse STFT back to output. The expression pedal controls the node position parameter that drives mask construction.

The Mask in Detail

The spectral mask is what makes this invention work. It is a frequency-domain gain function built from the detected fundamental frequency and the user's node position selection. Each surviving harmonic gets a narrow peak at its exact frequency; all other bins are attenuated to near silence.

Figure 2. Spectral masks for node positions 2, 3, and 4 applied to a guitar signal at A110 Hz. Each row shows the mask gain (0 to 1) across the frequency axis. Peaks appear only at the surviving harmonic frequencies. All other energy is suppressed to the floor value (~0.003).

Critical Implementation Constraint

The peak window half-width must be less than 40% of the inter-harmonic bin spacing, or adjacent mask peaks overlap and the filtering breaks down entirely. At 110 Hz with a 4096-sample window (10.77 Hz per bin), adjacent harmonics are separated by approximately 10 bins. The implementation uses 35% of this spacing as the mask half-width.

This constraint tightens at lower frequencies (bass guitar at 41 Hz is at the edge of reliable resolution with 4096-sample windows) and relaxes at higher frequencies. A bass-optimized implementation would use an 8192-sample window at the cost of higher latency (~186ms).

What It Sounds Like

These demonstrations use synthesized guitar audio (Karplus-Strong model at A110 Hz). The algorithm produces the same effect on real instrument recordings. Real guitar validation is in progress.

Figure 3. Spectrogram comparison: dry signal (top) vs. node 2 output (bottom). Odd harmonics are cleanly suppressed. Even harmonics are preserved at their original amplitude and phase relationships.

Figure 4. Spectrogram of the expression pedal sweep. As the node position changes over time, the surviving harmonics visibly shift. Watch harmonic lines fade in and out as the pedal moves from node 2 through node 6.

Dual-Bus Effects Demonstrations

The Platform Opportunity

The single-effect harmonic isolation is the entry point. The real invention is the dual-bus architecture it enables: separating any audio signal into a harmonics bus and a fundamental bus, routing each through independent effects chains, and recombining.

Once you can do that, every existing audio effect becomes two new effects: one applied to harmonics only, one applied to the fundamental only. Distortion, reverb, delay, compression, chorus, tremolo, filtering, panning, ring modulation — all become new effects when applied selectively to harmonic content. The combination space is essentially unlimited.

Figure 5. Dual-bus architecture. The spectral mask routes harmonic content to the harmonics bus and residual (fundamental + non-harmonic) content to the fundamental bus. Each bus passes through an independent effects chain before recombination. Both chains are user-configurable.

Fundamental Bus Derivation

The fundamental bus is derived by two equivalent methods: (a) applying the inverse mask (1 - M[k]) to the STFT and reconstructing, or (b) subtracting the harmonic bus output from the original dry signal. Method (b) is simpler to implement and has lower computational cost. Method (a) preserves better phase coherence in the residual.

Commercially Novel

No existing commercial product or published academic system implements this dual-bus harmonic separation architecture for real-time musical instrument processing with expression pedal control. The closest academic work (HPSS, SMS) targets offline analysis and does not produce the split-bus output structure described here.

What Is Claimed

A U.S. provisional patent application has been filed covering the following independent claims. The application further includes fourteen dependent claims specifying preferred embodiments. Claim 4 (dual-bus platform) is the broadest and most commercially significant claim.

Claim 1 (Method):
A method for processing an audio signal, comprising:
(a) receiving an audio input signal containing one or more harmonic
    partials of a fundamental frequency;
(b) detecting the fundamental frequency of the audio input signal;
(c) computing a frequency-domain representation of the audio input signal;
(d) constructing a spectral mask based on the detected fundamental
    frequency and a user-controlled node position parameter, wherein
    the spectral mask assigns a first gain value to frequency bins
    corresponding to harmonic partials that are integer multiples of
    a harmonic divisor determined by the node position parameter,
    and a second, lower gain value to other frequency bins;
(e) applying the spectral mask to the frequency-domain representation;
(f) reconstructing a time-domain output signal from the masked
    frequency-domain representation.

Claim 2 (Apparatus — Hardware Pedal):
An audio effects apparatus comprising:
(a) an audio input for receiving an audio signal from a musical instrument;
(b) an analog-to-digital converter;
(c) a digital signal processor configured to perform pitch detection,
    spectral analysis, spectral mask construction, mask application,
    and signal reconstruction as described in Claim 1;
(d) a digital-to-analog converter;
(e) an audio output;
(f) a user control interface comprising at least one of: a knob,
    an expression pedal input, or a footswitch;
wherein the user control interface controls the node position parameter
of the spectral mask.

Claim 3 (Software — Plugin):
A non-transitory computer-readable medium storing instructions that,
when executed by a processor, cause the processor to perform the method
of Claim 1, the instructions being packaged as an audio plugin
conforming to at least one of the VST3, Audio Unit, or AAX plugin
interface standards.

Claim 4 (Platform — Dual-Bus Selective Processing):
A method for selective audio processing, comprising:
(a) performing the method of Claim 1 to produce a harmonics signal;
(b) computing a fundamental signal by subtracting the harmonics signal
    from the original audio input signal, or by applying an inverse
    spectral mask;
(c) independently processing the harmonics signal through a first
    audio effects chain;
(d) independently processing the fundamental signal through a second
    audio effects chain;
(e) combining the independently processed harmonics signal and
    fundamental signal to produce a final output signal;
wherein the first and second audio effects chains apply different
audio processing operations.

Licensing and Partnership

Three paths are available depending on your use case and organization.

Current Status

A working Python prototype exists and has been validated on synthesized instrument audio. Real guitar recording validation is the immediate next step. The patent covers the algorithm independent of implementation language or platform. Patent filing establishes priority; hardware and plugin development can proceed in parallel.

Full Patent Application

Download the complete patent specification and figures as filed with the USPTO.

Get In Touch

If you have questions about the technology, are interested in licensing, or want to discuss partnership, reach out directly.

Or email directly: walter@sieveaudio.com