I am trying to use the timit time alignments to use to train model. I have been successful in training without alignments.
I tried changing the $CFG_FORCEDALIGN = 'no'; to 'yes'.
Is there any resource on providing me info on how to input the additional alignment information.
Thansk
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
You have to convert labels to sphinx3 phseg format and put them in lab or point to them with BW with -phsegdir option. The sample segmentation file:
SFrm EFrm SegAScr Phone
0 5 307602 SIL
6 35 1381919 SIL
36 46 86604 j
47 57 46263 aa
58 66 -15602 u
67 79 14704 zh
80 87 -61460 ee
88 98 276365 nn
99 102 28522 ee
103 107 -7615 f
108 127 559891 ss
128 139 -202429 ii
140 153 474894 l
154 156 -28983 ay
157 173 570128 h
174 202 1593092 SIL
Total score: 5023895
SegAScr is not important for training I suppose. But I'm not sure if it will have effect at all.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I am trying to use the timit time alignments to use to train model. I have been successful in training without alignments.
I tried changing the $CFG_FORCEDALIGN = 'no'; to 'yes'.
Is there any resource on providing me info on how to input the additional alignment information.
Thansk
You have to convert labels to sphinx3 phseg format and put them in lab or point to them with BW with -phsegdir option. The sample segmentation file:
SFrm EFrm SegAScr Phone 0 5 307602 SIL 6 35 1381919 SIL 36 46 86604 j 47 57 46263 aa 58 66 -15602 u 67 79 14704 zh 80 87 -61460 ee 88 98 276365 nn 99 102 28522 ee 103 107 -7615 f 108 127 559891 ss 128 139 -202429 ii 140 153 474894 l 154 156 -28983 ay 157 173 570128 h 174 202 1593092 SILTotal score: 5023895
SegAScr is not important for training I suppose. But I'm not sure if it will have effect at all.