Table 1 The computational complexity of feature extraction methods [15, 36]
Subset domain | Features (abbreviation) | Requires |
|---|---|---|
Time | Energy(g) | \(O\left( N \right)\) |
Zero-crossing rate (ZCR) | ||
Loudness (L) | ||
Frequency | Spectral flux (SX) | STFT which costs \(O\left( {N\log_{2} N} \right)\) |
Spectral roll-off (SR) | ||
Spectral flatness (SF) | ||
Wavelet | The mean of the subband’s coefficients (M) | 1-D Haar which costs \(O\left( N \right)\) |
The standard deviation of the subband’s coefficients (STD) | ||
The ratio of absolute mean values of two adjacent subbands (RM) | ||
The energy variance of the subband’s coefficients (E) | ||
The energy ratio between two adjacent subbands (ER) | ||
The temporal centroid per subband (TC) | ||
The difference between the temporal centroids of two adjacent subbands (TCD) | ||
The entropy per subband (P) |