Figure 2.

Estimated alignment accuracy by sound class by age for the five alignment algorithms and the marginal average over the five aligners. Lines represent the estimated population-average (fixed-effects) accuracy. Points represent the average accuracy for the classes of sounds for each child, so one point represents one child's accuracy for that sound class. Several key findings are visible here: (a) MFA-SAT was the most accurate overall; all of its age-trend lines are above 75%. (b) Vowels were the most accurately aligned sounds; the topmost age-trend line in each panel is the vowel line. (c) Fricatives were the sound class most affected by age: In every panel, there is a positive slope for the fricative age-trend line. The letters v, f, p, o are included to label the sound class for each line. MFA = Montreal Forced Aligner; SAT = speaker adaptive training; P2FA = Penn Phonetics Lab Forced Aligner.