AI Vocal Forensics Lab | Neural Vocoder vs Auto-Tune Acoustic Inspector

Forensic Context: Addresses listener controversies where heavily autotuned pop vocals are mistakenly labeled as "AI generated". Use the acoustic probe below to isolate the 2.4 kHz formant corridor and 8 kHz sibilance region to contrast hard pitch quantization with neural diffusion phase smearing.

Dual-Mode Spectrogram & Spectral Flux

PIPELINE: OVERPRODUCED DSP BAND: FULL SPECTRUM (20Hz - 20kHz) 60 FPS

0.00s

Forensic Bandpass Filter Isolator

Diagnostic Classifier

Forensic Assessment

Aggressive Studio DSP (Over-tuned)

Rigid pitch lock detected with zero retune transition latency. Fundamental harmonics remain phase-coherent with no neural diffusion smearing. Track shows classic pop comping and auto-tune artifacts, not generative AI.

Pitch Quantization Variance 1.2 cents

Phase Coherence Index 0.54

Formant Transition Roughness 0.78

Custom Audio Stem Dropzone

Upload Vocal Stem (WAV/MP3)

Client-side client decode & feature extraction. Audio is processed purely in your browser.

Loaded: Bundled Preset (Pony Pop Stem)

How to tell the difference:

• Auto-Tune / Melodyne: Instantaneous stair-step transitions between target semitones, but crystal-clear upper sibilance (>8 kHz) with intact phase relations.

• Neural Vocoders (Suno/Udio/DDSP): Mushy phase alignment, watery high-frequency flutter, and missing glottal pulses during consonant onsets.