The number of samples of each class of the observer 1 in DEV and EV subsets for IOR calculation.

In the DEV sub-dataset (n = 5 soundfiles), there are very few samples, i.e., 2 from the classes ‘c’ and ‘c4’, 4 from the class ‘c3’, 5 from the class ‘u’ and 9 from the class ‘h’, thus the results of these classes are not very reliable. We found similar results in the EV sub-dataset (n = 5 sound files) where there are very few samples from the classes ‘us’, ‘d’, ‘c’, and ‘c5’. (XLSX)