Figure 21.

Results of hyperparameter benchmarks for MFCC features. (a) shows a statistical analysis of the accuracy distributions for different hyperparameters, dependent on case and whether normalization was used for the filter-bank. (b) shows the accuracies reached as a function of hyperparameters $n_{\mathrm{mel}}$ and $\psi$, averaged across the tested window lengths for both pocket-case and hand-case.