Free Online Audio Harshness & Kurtosis Meter
Detect neural vocoder phase jitter, excessive sibilance and ear-fatiguing resonant peaks in your browser with 100% privacy. Built on sample-accurate higher-order spectral moments.
Drag & Drop audio file here or click to browse
Supports WAV, MP3, FLAC, OGG, M4A — processed 100% client-side in browser memory
Understanding Spectral Kurtosis & AI Audio Harshness
What is Excess Kurtosis in Audio?
Kurtosis is the fourth standardized moment of a signal's amplitude distribution. In natural acoustic music, audio waveforms follow an approximately Gaussian bell curve (excess kurtosis between 0.0 and 3.0). When an audio file exhibits kurtosis above 7.0 or 10.0, the signal contains abnormal sample-level spikes and non-Gaussian phase jitter that the human ear perceives as abrasive crunch or grating fatigue.
Why Do Suno & Udio Vocals Sound Scratchy?
Generative AI music models utilize neural vocoders (such as HiFi-GAN or BigVGAN) to reconstruct time-domain audio from latent diffusion tokens. At high frequencies above 5 kHz, these vocoders often struggle to maintain continuous sinusoidal phase coherence across consecutive frames. Instead of smooth harmonic overtones, they produce micro-phase jitter that creates abrasive sibilance and brittle high-frequency distortion.
Why Traditional EQs Fail on Phase Jitter
An equalizer scales signal amplitude linearly: reducing 10 kHz by -4 dB lowers both the musical vocal tone and the vocoder phase noise equally. The signal-to-noise ratio within the band remains identical, leaving the vocal dull yet still scratchy. True restoration requires sample-level slew-rate limiting, phase-coherent dispersion or Mid/Side-targeted processing.
Frequently Asked Questions
Are my audio files uploaded to a remote server?
No. All file decoding, 6-band filtering and kurtosis calculations run 100% locally in your web browser using the Web Audio API. Your tracks never leave your computer or device.
What is an acceptable kurtosis value for commercial masters?
For professional pop, rock and electronic masters, excess kurtosis in the presence and sibilance bands (2 kHz to 8 kHz) should ideally sit between 1.5 and 5.0. Values consistently exceeding 8.0 in the Air band indicate neural vocoder artifacts or harsh digital clipping.
How does Mixify fix detected AI vocal harshness?
Mixify Studio uses zero-phase Butterworth decimation, neural AFFTDN denoising, sample-level slew-rate soft rounding and Mid/Side Wolfson groove decorrelation to eliminate phase jitter while preserving commercial top-end air.