Hello,
I noticed that one of my (sphinx3 in flatfwd mode) lattice showed following edges
Nodes 14 (NODEID WORD STARTFRAME FIRST-ENDFRAME LAST-ENDFRAME) 0 92 94 96 1 +BABBLE+ 85 87 96 2 bhaat 80 92 96 3 +BG_NOISE+ 73 75 91 4 +AIR+ 73 91 96 5 67 69 93 6 <sil> 67 69 93</sil> 7 +AIR+ 63 65 90 8 bhaat 40 63 84 9 +BABBLE+ 8 10 14 10 bhaat 4 12 25 11 baajarii(2) 4 92 96 12 ananasa 3 35 96 13 0 2 17
Both /s and SIL have the same acoustic scores over the range.
Shouldn't /s have been deleted from the lattice knowing that it doesn't end in the last frame?
According to http://www.cs.cmu.edu/~archan/s_info/Sphinx3/doc/s3_description.html#sec_wordlat
BestSegAscr section, a historical remnant now essentially empty
Could someone please confirm that, that info is outdated, in the sense that the scores given in the lattice are 'reliable' (but un-normalized) scores?
Thanks.
Log in to post a comment.
Hello,
I noticed that one of my (sphinx3 in flatfwd mode) lattice showed following
edges
Nodes 14 (NODEID WORD STARTFRAME FIRST-ENDFRAME LAST-ENDFRAME)
0 92 94 96
1 +BABBLE+ 85 87 96
2 bhaat 80 92 96
3 +BG_NOISE+ 73 75 91
4 +AIR+ 73 91 96
5 67 69 93
6 <sil> 67 69 93</sil>
7 +AIR+ 63 65 90
8 bhaat 40 63 84
9 +BABBLE+ 8 10 14
10 bhaat 4 12 25
11 baajarii(2) 4 92 96
12 ananasa 3 35 96
13
0 2 17Both /s and SIL have the same acoustic scores over the range.
Shouldn't /s have been deleted from the lattice knowing that it doesn't end in
the last frame?
According to http://www.cs.cmu.edu/~archan/s_info/Sphinx3/doc/s3_description.html#sec_wordlat
BestSegAscr section, a historical remnant now essentially empty
Could someone please confirm that, that info is outdated, in the sense that
the scores given in the lattice are 'reliable' (but un-normalized) scores?
Thanks.