CORIS.THIRI.AI / EXPERIMENTS / WIDGET 02

Voice-Adaptive Audio Bed Ducking & Spectral Carving

Standard AI Failure
No Ducking (Vocal Masking)

Music bed clashes in the 1–3 kHz formant pocket, muddying spoken words.

Naive DAW Ducking
Flat Volume Ducking (-12 dB)

Pumps entire master track down, killing sub-bass punch and rhythm drive.

THIRI Architecture
Spectral Notch + Smooth Sidechain

Surgically carves 1.8 kHz pocket while preserving 100% sub-bass energy.

FFT SPECTRUM ANALYSIS: Sub (40Hz) → Vocal Formant Pocket (1.8kHz) → Air (10kHz+) STATUS: VOCAL MASKING DETECTED
Formant Pocket Frequency: 1.85 kHz Center of human voice intelligibility band
Spectral Notch Depth: -6.0 dB Depth of parametric EQ carve under vocal bursts
Sidechain Release Time: 250 ms Envelope recovery time after voice ceases
Speech Intelligibility (SII) 0.42 POOR
Formant Masking +8.4 dB
Sub-Bass Energy Retained 100% FULL
DSP State CLASHING
👤 Non-Tech / Creator Experience Interactive Ear X-Ray
The "Cookie-Cutter Keyhole" Ear Test

Everyday Analogy: Turning down the whole music track when someone speaks is like dimming all the lights in a stadium just to read a menu. Instead, THIRI uses an acoustic cookie-cutter at 1.85 kHz: it carves a tiny keyhole just for the human voice while keeping 100% of the deep bass and drum energy pumping.

🎧 Click below to isolate what gets removed vs what stays:
💻 Developer / Live DSP Filtergraph
WebAudio Biquad + FFmpeg Sidechain EQ

Live parametric peaking filter & sidechain compressor generated from your slider values: