When doing a phone only alignment and a word only alignment and comparing the results, some words contain silences.
Example: the word "suite" should end at 1.6475, not 2.0575.
F01_R_B1_C1_0 1.0975 2.0575 suite
F01_R_B1_C1_0 1.0975 1.2375 1.0000 s
F01_R_B1_C1_0 1.2375 1.3075 1.0000 y
F01_R_B1_C1_0 1.3075 1.3875 1.0000 i
F01_R_B1_C1_0 1.3875 1.6475 1.0000 t
F01_R_B1_C1_0 1.6475 2.0575 1.0000 SIL
To reproduce:
Using the same language model, acoustic model, features and corpus, do
abkhazia align --phones-only $corpus -o $output_dir/alignment_phone \ -l $output_dir/language -a $output_dir/am_trisa -f $output_dir/features
and
abkhazia align --words-only $corpus -o $output_dir/alignment_words \ -l $output_dir/language -a $output_dir/am_trisa -f $output_dir/features
When doing a phone only alignment and a word only alignment and comparing the results, some words contain silences. Example: the word "suite" should end at 1.6475, not 2.0575.
To reproduce: Using the same language model, acoustic model, features and corpus, do
abkhazia align --phones-only $corpus -o $output_dir/alignment_phone \ -l $output_dir/language -a $output_dir/am_trisa -f $output_dir/featuresandabkhazia align --words-only $corpus -o $output_dir/alignment_words \ -l $output_dir/language -a $output_dir/am_trisa -f $output_dir/featuresand compare the results