ScreamGate

ScreamGate for Windows

Portable Windows package / realtime audio gate / spectrum targeting

ScreamGate delays output by a tiny look-ahead buffer, analyzes incoming audio, and drops gain when a targeted high-frequency spike appears. It is built for people who want a practical YouTube volume filter that can make everything ASMR-calm without crushing ordinary dialogue.

The current default build starts simple: one calibration flow, a visible countdown, an early Arm Now button, live lock readouts, collapsed advanced tuning, hard dB capping, and a scream fingerprint detector for abrupt chaotic presence-band bursts.

Calibration WorkflowLive IntensityPortableYouTube Gate

Start Calibration

Starts audio, clears old memory, and learns the loudest normal dB peak as the ceiling.

Arm Now

Stops calibration early and arms the gate using whatever normal baseline has already been heard.

New Video

Click Restart Calibration to clear the previous scene memory and recalibrate from scratch.

Lock Readout

Once armed, ScreamGate reports the dB and frequency it locks onto, including hard kills in the 1000-2000 Hz zone.

Scream fingerprint

Looks for abrupt attack, high presence-band energy, spectral flux, and roughness in 1000-4000 Hz so it can catch more than one exact scream shape.

Hard cap dB

Defaults to -45.5 dB and can be moved live. Audio below that line is protected from suppression.

Drag across the spectrum, then pick the painful points.

While audio is playing, move or drag your mouse over the spectral graph to inspect the exact frequency and level under the cursor. Low bass lives on the left; sharper high-pitched screams usually appear farther to the right.

Click the spike area you want to cut out. The click is dual-axis: left/right chooses the frequency, up/down chooses the dB floor. ScreamGate uses that selected spectral point to target a narrow frequency band, instead of muting the whole movie just because something else is loud.

The target detector compares that band against its own learned baseline. That means louder bass can pass, while a smaller but abnormal high-frequency scream band can still get clamped.

Automatic gate: after calibration, ScreamGate learns the loudest spectrum and densest normal waveform it has already heard, then clamps new high-frequency bursts or tight waveform walls that jump above that history.

Pause-to-target: when playback goes quiet, the spectrum holds the last active frame so you can click the painful area instead of watching it vanish.

Graph legend

Green is live audio. Blue is the learned baseline. Orange is the session max. Gray is the session min. The red vertical line marks target frequency, the yellow horizontal line marks target dB, and the yellow dot marks the exact clicked point. A DENSE label means the waveform-density detector caught a scream-like wall of sound.

Target width Hz

Narrow catches a precise scream band. Wider catches nearby harmonics but may mute more content.

Target over baseline dB

Lower values trigger faster. Higher values reduce false positives.

Intensity

Intensity stays usable after calibration. It shifts sensitivity and clamp strength without erasing the learned baseline.

Look-ahead / hold / release

Look-ahead catches spikes before output, hold keeps the gate down briefly, and release fades sound back in smoothly.

What the settings mean

The app keeps these controls collapsed by default. Look-ahead is the small safety delay that lets ScreamGate react before a spike reaches your speakers. Hold keeps reduction active briefly, release controls how smoothly sound returns, and attenuation is how much a caught spike gets reduced. In plain English: calibrate normal playback, then let the gate catch new spectral points that jump above normal.

Executive Summary: Spectral Probability Density in Adaptive Acoustic Filtering

This paper details a paradigm shift in real-time digital signal processing, moving away from traditional deterministic amplitude thresholding toward a framework rooted in statistical probability. By treating the acoustic spectrum as a dynamic probability density function, ScreamGate distinguishes harmless conversational transients from sustained acoustic anomalies.

1. The Core Innovation: Wavelength as Probability

Traditional DSP limiters analyze audio frame by frame. A brief speech syllable, such as a plosive P or T sound, can share the same instantaneous frequency and amplitude as a loud scream, causing ordinary limiters to clamp normal dialogue.

ScreamGate adds temporal context. Instead of measuring only absolute volume in dBFS, the algorithm evaluates spectral energy across a sliding look-ahead window.

Transient identification: high-amplitude spikes lasting less than about 0.1 seconds produce low probability density. The system treats these as safe linguistic transients and applies unity gain, letting speech pass untouched.

Resonance identification: when energy continuously saturates a frequency band over consecutive frames, probability density approaches 100%. The system treats this as a sustained resonant threat and applies localized attenuation to the target bins.

The architecture uses a Short-Time Fourier Transform pipeline with a Hann window and 75% overlap-add topology to reduce artifacts such as clicking, popping, and phase smearing.

2. Primary Applications

Consumer media optimization: the ScreamGate paradigm targets highly dynamic multimedia content such as horror and action cinema, suppressing unexpected high-roughness vocal screams without muddying dialogue, background score, or ambience.

Next-generation digital ear protection: in tactical headsets or industrial active hearing protection, the same per-bin probability mask could suppress a hazardous shriek while preserving situational awareness.

Selective acoustic shielding: instead of muting the entire environment during a loud event, the system targets the specific persistent frequency of an industrial shriek or hazard.

Continuous communication: ordinary conversation, especially low-frequency speech content around 200-500 Hz, can remain unattenuated when it does not show the probability profile of sustained danger.

3. Secondary Potential Applications

Live broadcast and telephony: automatic clamping of microphone feedback loops, audience spikes, or line noise during broadcasts and VoIP calls without interrupting the speaker's signal.

Acoustic drone and UAV detection: a probability density matrix can isolate the persistent harmonic whine of rotors from chaotic environmental noise such as wind or traffic.

Smart home and nursery monitoring: brief household transients like dropped objects or door slams can be ignored, while sustained distress signals such as crying or alarms remain detectable.