Monophonic Noise Filter

Parameters

Tolerance — How close to a detected harmonic frequency a spectral bin must be (in Hz) to be classified as “harmonic” rather than “noise”, 5–200. Tight tolerance (5–20 Hz) keeps only bins very close to integer multiples of the detected pitch — a strict harmonic mask. Loose tolerance (50–200 Hz) accepts more bins around each harmonic — preserves vibrato, breath, and natural variation. Tighter values produce cleaner harmonic separation; looser values keep the source sounding natural.

Mix — Wet/dry blend, 0–100%. 100% is fully filtered output; lower values blend the original signal back in.

Softness — How abrupt the harmonic-vs-noise mask is, 0–1. 0 is a hard binary mask (each bin is either harmonic or noise — can produce artifacts on transitions). 1 is a fully smoothed mask with gradual transitions between regions (more natural, fewer artifacts, but less surgical). 0.3–0.6 is balanced for most use cases.

Gain — Output makeup gain in dB, −24 to +24. Compensates for level changes after extraction.

Additional controls

Mode — Selects which output you keep. Noise Only — outputs the noise component (everything that isn’t the detected harmonic series). Harmonics Only — outputs the harmonic component (the detected pitch and its overtones). Both Outputs — exposes two output ports; harmonics on output 0, noise on output 1, summing back to the original. Use Both when you want to process each component independently in the graph and recombine downstream.

N Harmonics — Number of harmonics tracked above the fundamental, 4–64. More harmonics capture brighter sources (instruments with rich overtones); fewer harmonics are sufficient for darker sources (bass, voiced speech). 16–32 is balanced for most monophonic sources.

Min F0 / max F0 — Lower and upper bounds (in Hz) for fundamental frequency detection. Constrains where the pitch tracker hunts. Set min above the lowest expected pitch and max below the highest. For male voice, ~80–400 Hz; for female voice, ~150–600 Hz; for solo instruments, set to the instrument’s range. Tightening the range improves tracking accuracy.

Confidence — Minimum pitch-detection confidence (0.1–0.9) required to consider the source as having a stable pitch. Below this threshold the node assumes “no pitched content” and treats the frame as noise. Higher confidence requirements produce cleaner results on stable pitched material; lower requirements help with breathy, noisy, or partially-pitched sources.

Median — Frame window size (1–15) for median filtering of detected pitch. Smooths short-term pitch detection errors without over-smearing legitimate pitch movements (vibrato, slides). 3–5 is balanced; higher values produce smoother but slower-tracking pitch.

Smooth — Pitch smoothing alpha, 0–0.95. Higher values smooth pitch detection over time more aggressively. Use sparingly — too much smoothing makes the tracker miss fast pitch changes (slides, ornaments).

About Monophonic Noise Filter

This node assumes the input is monophonic — a single voice, a single solo instrument — and uses pitch detection to identify which spectral content belongs to that pitch’s harmonic series and which is “noise” (breath, bow noise, hiss, room tone, transients). Once classified, the two are separable. Use it to: clean up vocal recordings (extract harmonics, discard breath); isolate solo instrument pitch from background bleed; route harmonic content separately from noisy components for independent processing. It will not work on chords or polyphonic material — for that, the Spectral Compressor or other multiband tools are more appropriate. The “Both Outputs” mode is the most flexible: extract once, process each component separately in the graph, and decide later how to recombine them.

Generated 2026-05-05 from K2K_Dev@96730bdc by scripts/gen_lexique.py. Edit _intros/ or _overrides/, not this file.