Spectral Peak Extractor

Parameters

Peak Width — Bandwidth around each detected peak that gets included in the extraction, in Hz, 10–200. Tight widths (10–30 Hz) capture only the peak itself — surgical isolation, useful for grabbing a single resonance. Wider (100–200 Hz) include surrounding spectrum — softer extraction with more body around each peak.

Threshold — Minimum peak level (in dB) required to be considered, −60 to 0. Higher (less negative) values catch only the most prominent peaks (formants, dominant overtones); lower values catch subtler peaks too (lighter overtones, lower-level resonances). −40 dB is a good starting point.

Additional controls

Peaks — Maximum number of peaks to extract, 1–20. Sets the upper limit on simultaneous peak tracking. Lower (1–5) for surgical isolation of the strongest resonances (formants, fundamentals + first few harmonics); higher (10–20) for capturing rich harmonic textures and complex timbral content.

Time Avg — Time-averaging method for stable peak detection across frames:

• Mean — average peak detection across the analysis window. Smooth, slow-tracking peak set; good for sustained tones.
• Median — median-based detection; robust to single-frame anomalies. Best for material with brief transient noise that shouldn’t be classified as peaks.
• Max — uses the strongest peak detected in each window. Reactive, catches brief peaks; more variation frame-to-frame.

For sustained tones, Median or Mean; for percussive/transient-rich material, Max.

About Spectral Peak Extractor

This node finds the loudest peaks in the spectrum — typically the resonances and overtones that define a sound’s character. Each peak gets a window of width peak_width around it; the extracted output keeps only those windows, the rejected output keeps everything else. Use it for: isolating formants and resonances for processing, extracting just the harmonic skeleton of a complex sound, building “skeletal” versions of recordings (just the strongest spectral content), or as a pre-processing step before pitch detection. Compare with the Formant Extractor which uses cepstral analysis to find true vocal-tract formants — the Peak Extractor is more general (any spectral peak qualifies) and works on any source, but doesn’t distinguish formants from random resonant peaks the way Formant Extractor does.

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