dear povey，
now i have a problem about the fst.i read fst.30.gz and read it in fstcopy. i am not sure the first and second rows is the trans-id.and the forth is the word-id.i do not understand the third row is what.if the first and second rows is state id,i am sure the third row is trans-id and the forth is word-id.but in our timit recipe in kaldi,it only have 48 phones and i see the si2022 have 20 phones,and the hmm trans-id is 6,so maybe trans-id.so i do not what means.thank you for your help.
best wishes,
ben
It looks like this is the acceptor format of OpenFst. The 3rd field
is the word-id and the last field is the cost (negated
log-likelihood), coming from the lexicon (pron-prob of
silence/not-silence).
Dan
dear povey，
now i have a problem about the fst.i read fst.30.gz and read it in fstcopy.
i am not sure the first and second rows is the trans-id.and the forth is the
word-id.i do not understand the third row is what.if the first and second
rows is state id,i am sure the third row is trans-id and the forth is
word-id.but in our timit recipe in kaldi,it only have 48 phones and i see
the si2022 have 20 phones,and the hmm trans-id is 6,so maybe trans-id.so i
do not what means.thank you for your help.
best wishes,
ben
dear povey,
thank you for your reply.but it have five rows.the fifth rows is cost,but always
empty.the forth is word-id.because in timit,the word.txt is the same as the lexcious.txt and phone.txt,and it only no more than 50.so i am not sure what is it.
thank you for your reply again.
best wishes,
ben
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
i know it .but in timit,every HMM is have 3 states.and this utterance is 20 phones.so the first and second rows should be no more than 60,but the first and second rows is 120 now.so i do not know that.
thank you for your reply.
ben
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
dear povey，
now i have a problem about the fst.i read fst.30.gz and read it in fstcopy. i am not sure the first and second rows is the trans-id.and the forth is the word-id.i do not understand the third row is what.if the first and second rows is state id,i am sure the third row is trans-id and the forth is word-id.but in our timit recipe in kaldi,it only have 48 phones and i see the si2022 have 20 phones,and the hmm trans-id is 6,so maybe trans-id.so i do not what means.thank you for your help.
best wishes,
ben
faem0_si2022
0 1 2 0
1 2 4 0
1 1 1 0
2 3 6 0
2 2 3 0
3 4 2 38 0.693359
3 117 266 38 0.693359
3 3 5 0
4 115 4 0
4 4 1 0
5 6 268 0
6 7 270 0
6 6 267 0
7 8 2 0 0.693359
7 9 20 3 0.693359
7 7 269 0
8 113 4 0
8 8 1 0
9 10 22 0
9 9 19 0
10 11 24 0
10 10 21 0
11 12 2 0 0.693359
11 13 80 13 0.693359
11 11 23 0
12 111 4 0
12 12 1 0
13 14 82 0
13 13 79 0
14 15 84 0
14 14 81 0
15 16 2 0 0.693359
15 17 32 5 0.693359
15 15 83 0
16 109 4 0
16 16 1 0
17 18 34 0
17 17 31 0
18 19 36 0
18 18 33 0
19 20 2 0 0.693359
19 21 122 20 0.693359
19 19 35 0
20 107 4 0
20 20 1 0
21 22 124 0
21 21 121 0
22 23 126 0
22 22 123 0
23 24 2 0 0.693359
23 25 146 24 0.693359
23 23 125 0
24 105 4 0
24 24 1 0
25 26 148 0
25 25 145 0
26 27 150 0
26 26 147 0
27 28 2 0 0.693359
27 29 62 10 0.693359
27 27 149 0
28 103 4 0
28 28 1 0
29 30 64 0
29 29 61 0
30 31 66 0
30 30 63 0
31 32 2 0 0.693359
31 33 68 11 0.693359
31 31 65 0
32 101 4 0
32 32 1 0
33 34 70 0
33 33 67 0
34 35 72 0
34 34 69 0
35 36 2 0 0.693359
35 37 242 41 0.693359
35 35 71 0
36 99 4 0
36 36 1 0
37 38 244 0
37 37 241 0
38 39 246 0
38 38 243 0
39 40 2 0 0.693359
39 41 224 37 0.693359
39 39 245 0
40 97 4 0
40 40 1 0
41 42 226 0
41 41 223 0
42 43 228 0
42 42 225 0
43 44 2 0 0.693359
43 45 152 25 0.693359
43 43 227 0
44 95 4 0
44 44 1 0
45 46 154 0
45 45 151 0
46 47 156 0
46 46 153 0
47 48 2 0 0.693359
47 49 260 44 0.693359
47 47 155 0
48 93 4 0
48 48 1 0
49 50 262 0
49 49 259 0
50 51 264 0
50 50 261 0
51 52 2 0 0.693359
51 53 68 11 0.693359
51 51 263 0
52 91 4 0
52 52 1 0
53 54 70 0
53 53 67 0
54 55 72 0
54 54 69 0
55 56 2 0 0.693359
55 57 212 35 0.693359
55 55 71 0
56 89 4 0
56 56 1 0
57 58 214 0
57 57 211 0
58 59 216 0
58 58 213 0
59 60 2 0 0.693359
59 61 44 7 0.693359
59 59 215 0
60 87 4 0
60 60 1 0
61 62 46 0
61 61 43 0
62 63 48 0
62 62 45 0
63 64 2 0 0.693359
63 65 254 43 0.693359
63 63 47 0
64 85 4 0
64 64 1 0
65 66 256 0
65 65 253 0
66 67 258 0
66 66 255 0
67 68 2 0 0.693359
67 69 122 20 0.693359
67 67 257 0
68 83 4 0
68 68 1 0
69 70 124 0
69 69 121 0
70 71 126 0
70 70 123 0
71 72 2 0 0.693359
71 73 26 4 0.693359
71 71 125 0
72 81 4 0
72 72 1 0
73 74 28 0
73 73 25 0
74 75 30 0
74 74 27 0
75 76 2 0
75 75 29 0
76 77 4 0
76 76 1 0
77 78 6 0
77 77 3 0
78 79 2 38 0.693359
78 118 0 38 0.693359
78 78 5 0
79 80 4 0
79 79 1 0
80 119 6 0
80 80 3 0
81 82 6 0
81 81 3 0
82 73 26 4
82 82 5 0
83 84 6 0
83 83 3 0
84 69 122 20
84 84 5 0
85 86 6 0
85 85 3 0
86 65 254 43
86 86 5 0
87 88 6 0
87 87 3 0
88 61 44 7
88 88 5 0
89 90 6 0
89 89 3 0
90 57 212 35
90 90 5 0
91 92 6 0
91 91 3 0
92 53 68 11
92 92 5 0
93 94 6 0
93 93 3 0
94 49 260 44
94 94 5 0
95 96 6 0
95 95 3 0
96 45 152 25
96 96 5 0
97 98 6 0
97 97 3 0
98 41 224 37
98 98 5 0
99 100 6 0
99 99 3 0
100 37 242 41
100 100 5 0
101 102 6 0
101 101 3 0
102 33 68 11
102 102 5 0
103 104 6 0
103 103 3 0
104 29 62 10
104 104 5 0
105 106 6 0
105 105 3 0
106 25 146 24
106 106 5 0
107 108 6 0
107 107 3 0
108 21 122 20
108 108 5 0
109 110 6 0
109 109 3 0
110 17 32 5
110 110 5 0
111 112 6 0
111 111 3 0
112 13 80 13
112 112 5 0
113 114 6 0
113 113 3 0
114 9 20 3
114 114 5 0
115 116 6 0
115 115 3 0
116 120 266 45
116 116 5 0
117 5 0 45
117 117 265 0
118
119 118 0 0
119 119 5 0
120 5 0 0
120 120 265 0
It looks like this is the acceptor format of OpenFst. The 3rd field
is the word-id and the last field is the cost (negated
log-likelihood), coming from the lexicon (pron-prob of
silence/not-silence).
Dan
On Fri, Nov 28, 2014 at 7:31 AM, wbgxx333 wbgxx333@users.sf.net wrote:
dear povey,
thank you for your reply.but it have five rows.the fifth rows is cost,but always
empty.the forth is word-id.because in timit,the word.txt is the same as the lexcious.txt and phone.txt,and it only no more than 50.so i am not sure what is it.
thank you for your reply again.
best wishes,
ben
Oh, I see, this is a per-utterance decoding graph.
The inputs are transition-ids.
Dan
yes,it is a per-utterance decoding graph.you means the third rows is transition-id?and the first and second rows is also transition-id?
ben
First and second rows are begin/end state in the FST; see
www.openfst.org to understand the FST format.
Dan
On Fri, Nov 28, 2014 at 10:20 PM, wbgxx333 wbgxx333@users.sf.net wrote:
i know it .but in timit,every HMM is have 3 states.and this utterance is 20 phones.so the first and second rows should be no more than 60,but the first and second rows is 120 now.so i do not know that.
thank you for your reply.
ben