Table 2 Paradigm-wise (a)–(d) best system configuration regarding audio normalisation, feature normalisation, and training partition upsampling, as well as respective classification results in form of class-specific numbers of (in)correctly assigned vocalisations (confusion matrices) and unweighted average recall (UAR) for both vocalisation-wise and infant-wise evaluation scenario.

From: Automatic vocalisation-based detection of fragile X syndrome and Rett syndrome

  1. UAR values are rounded to three decimal places. AD, atypical development; C, support vector machine kernel complexity parameter; FXS, fragile X syndrome; n, number of participants; RTT, Rett syndrome; TD, typical development; ♀, female;  ♂, male; colour code for confusion matrices: greyscale proportional to the class-specific number of vocalisations between white = 0 % of class-specific vocalisations and black = 100 % of class-specific vocalisations.
Back to article page