Feature (abbreviation) | Description | # of variables |
---|---|---|
Spectral centroid (SC) | Center of mass of the spectrum | 1 |
Spectral rolloff point (SR) | Right skewness of the power spectrum | 1 |
Spectral flux (SF) | Amount of spectral change of the signal | 1 |
Compactness | Sum of results of fast Fourier transform over frequency bins | 1 |
Spectral variability (SV) | Variance of the magnitude spectrum | 1 |
Root mean square (RMS) | Power of the signal | 1 |
Fraction of low energy windows (FLEW) | Quietness of the signal relative to the rest of the signal | 1 |
Zero crossings (ZC) | The number of times the signal changes sign from one sample to another | 1 |
Strongest beat (SB) | Highest bin in the beat histogram | 1 |
Beat sum (BS) | Sum of all values in the beat histogram | 1 |
Strength of strongest beat (SSB) | Strength of the strongest beat in the signal | 1 |
Strongest frequency via ZC (SF-ZC) | Strongest frequency in the signal by looking at the ZC | 1 |
Strongest frequency via SC (SF-SC) | Strongest frequency in the signal by looking at the SC | 1 |
Strongest frequency via FFT max (SF-FFT) | Highest bin in the power spectrum | 1 |
MFCC | Short-term power spectrum based on the nonlinear mel scale of frequency | 13 (0–12) |
Constant-Q based MFCC (CQ-MFCC) | MFCC that directly calculates the logarithmic frequency bins | |
Linear predictive coding (LPC) | Spectral envelope based on the information of a linear predictive model | 10 (0–9) |
Method of moments (MM) | Calculation of the first 5 statistical method of moments | 5 (0–4) |
Relative difference function (RDF) | Log of the derivative of the RMS | 1 |
Area method of moments (AMoM) | Numeric quantities at some distance from a reference point or axis | 10 (0–9) |
AMoM of MFCC | AMoM derived with MFCC values instead of the density distribution function | 10 (0–9) |
AMoM of CQ-MFCC | AMoM derived with CQ-MFCC values instead of the density distribution function | 10 (0–9) |
AMoM of Log of CQ Transform (LCQT) | AMoM derived with Log Constant-Q Transform values instead of the density distribution function | 10 (0–9) |