Figure 6
From: Analytical Features: A Knowledge-Based Approach to Audio Feature Generation
(a) A performance comparison of analytical versus reference features, for various feature set dimensions. AFs perform better, and the improvement is more important for feature sets of small dimension. (b) A comparison of various regression classifiers to approximate the "spectral compactness" feature. The best AF yields a performance of 85%. It outperforms (1) the best reference feature (43%); (2) a feature set with 6 general spectral features (19%); (3) the 10 first MFCC (38%); and (4) the whole set of reference features of dimension 100 (83%). A set with the 100 best AFs further improves the correlation up to 90%.