Audio Sources

Load one song file. Separation executes locally via browser FFT.

Select Song File

Load a song file to begin.

Separating Vocals and Instruments... 0%

Engine V8 — VAD-Guided Separation + Spectral Inpainting: A Voice Activity Detection pre-pass now scans the entire track before any FFT work begins. It uses RMS energy, spectral centroid (300–4200Hz vocal register), and zero-crossing rate to score each frame from 0 (pure music) to 1 (singing). The ICC vocal mask is multiplied by this VAD score — so in instrumental sections the mask collapses to zero and the music track is completely untouched. After separation, Spectral Inpainting averages the music spectrum from surrounding clean frames and blends it back into the gaps where vocals were removed, reconstructing continuous texture.

Separation Tuning Controls

Adjust how aggressively the separation engine cuts or isolates audio elements.

🎤 Vocal Isolation Settings Decide how tightly centered and isolated you want the vocal track to sound. Vocal Extraction Sharpness 60.0 Higher values aggressively throw out instruments, leaving only perfectly centered dry vocals. Silence Background Noise (Gate) -55 dB Mutes the vocal track during quiet moments to clear out background instrument hiss. Filter Out Extreme Pitches for Vocals 110Hz - 8500Hz

Use left slider to ignore deep instrument bass (low pitch) and right slider to ignore high sizzle/cymbals (high pitch).

🎸 Instrumental Track Settings (Vocal Removal) Decide how cleanly and deeply vocals are carved out from the music. Vocal Removal Depth 60.0 Controls how much of the estimated vocal is subtracted from the music (0=none, 150=100%). Bass, kick, and cymbals are never affected regardless of this setting — they fall outside the vocal frequency mask. Spectral Inpainting Depth (Heal Music Gaps) 80% After removing vocals, this blends in music energy averaged from surrounding non-vocal frames to fill the spectral hole. 0% = raw subtraction only (may sound slightly hollow). 80–100% = smooth, continuous music with the vocal gap healed.

⚡ Quick Automatic Presets Sets the optimal sliders for your goal instantly. The manual sliders will update and add on top of these automatic edits.