Fig. 9
From: Free resources for forced phonetic alignment in Brazilian Portuguese based on Kaldi toolkit
Example of alignments for the utterance “ela tem muita fome.” Ground-truth boundaries are shown in green, while automatic ones produced by Kaldi’s monophone- and TDNN-F-based models are shown in blue and orange, respectively. Tokens (phonemes) are shown at the bottom at their respective in-between-boundaries middle position. Notice that phonemes /j\(\sim\):m/ and /j\(\sim\):n/ have been created in the ground-truth through a merge of the individual phones /j\(\sim\), /m/ and /n/, based on M2M mappings