FeedbackGuard vs Spectral Hold

FeedbackGuard is an Audio Unit v3 effect that stops acoustic feedback before it ruins your performance. Whether you are singing live, speaking on stage, or recording vocals on iPad, FeedbackGuard listens for runaway howling and automatically places precise notch filters — so you can focus on the show, not the squeal. HOW IT WORKS FeedbackGuard uses PNPR (Peak-to-Neighbourhood Power Ratio) analysis combined with a growth check to tell real feedback apart from normal voice, speech, or room tone. When a howling peak starts to build, the plugin tracks it with parabolic peak detection and deploys an adaptive notch bank that reacts in milliseconds. Designed for live use: attack and release times, suppression depth, and notch count are all under your control. KEY FEATURES • Real-time howling detection (PNPR + run-up check) • Adaptive notch bank — up to 24 simultaneous notches • Live spectrum display with active notch markers • "Balanced" preset for safe, transparent defaults • 3 user preset slots — tap to recall, hold to save • Fine-tuned controls: Sensitivity, Notch Threshold, Suppression, Q, Attack, Release • Optional harmonic tracking for stubborn feedback tones • Optional spectral noise reduction for quiet passages • Full bypass GET STARTED IN SECONDS 1. Open your AUv3 host (GarageBand, AUM, Cubasis, and more) 2. Insert FeedbackGuard as an audio effect on your vocal or monitor channel 3. Tap "Balanced" for recommended live settings 4. Raise Notch Threshold if notches appear on voice or room tone IDEAL FOR • Live vocals and speech • Rehearsals, worship, small PA setups • Mobile recording rigs on iPhone and iPad • Any situation where monitor feedback is a risk COMPATIBILITY • Audio Unit v3 (aufx) effect • iPhone and iPad • Works in any AUv3-compatible host • Mono or stereo, sample rates from 8 kHz to 192 kHz FeedbackGuard — Adaptive Feedback Suppressor by Norwin Palme.

SPECTRAL HOLD A spectral processor that combines a frequency-domain tilt EQ with a spectral freeze. Audio is analyzed with a short-time Fourier transform (1024-point FFT, 75% overlap, Hann window). Each frame's bins can be tilted around a pivot frequency, and the most recent frame can be frozen and held indefinitely, turning any moment of the input into a sustained spectral drone. ------ CONTROLS TILT Applies a spectral tilt around the pivot frequency. The gain of each bin is scaled by tilt * 3 dB per octave of distance from the pivot. Negative values darken the sound by attenuating highs and boosting lows; positive values brighten it by boosting highs and attenuating lows. At 0 the spectrum passes unchanged. Range: -1.00 to +1.00 (so +/-3 dB/octave at the extremes). PIVOT Sets the frequency around which the tilt rotates. Bins above the pivot are affected in one direction and bins below in the other; the pivot itself is left at unity gain. Lower the pivot to pivot the tilt around the low end, raise it to rotate around the highs. Range: 20 Hz to 20 kHz (exponential), automatically clamped below the host Nyquist frequency. FREEZE Captures and holds the most recently analyzed spectral frame. While engaged, the stored magnitude and phase of every bin are replayed on each hop instead of the live input, producing a continuous frozen texture. Tilt and pivot remain active on the held spectrum, so the frozen sound can still be shaped. Disengage to return to live processing. DRY/WET Blends the unprocessed (dry) input with the processed (wet) output. At 0% only the dry signal is heard; at 100% only the wet. The dry path is internally delay compensated so it stays time- and phase-aligned with the wet path at every mix setting, with no comb filtering. Useful for layering a frozen pad underneath the live signal, or easing in a tilt. ------ NOTE Because the effect works frame by frame, it reports a latency of one FFT frame (1024 samples, about 21 ms at 48 kHz) to the host for accurate delay compensation.