when load the model, Null point always returned. about shiro HOT 3 CLOSED

Hello. How did you create flat.hsmm? Would you share the file for an inspection?

from shiro.

Hello. How did you create flat.hsmm? Would you share the file for an inspection?

I just follow up your kind instruction.
like this
lua shiro-mkseg.lua index.csv \ -m phonemap.json \ -d "../cmu_us_bdl_arctic/orig/" \ -e .param -n 36 -L sil -R sil > unaligned-segmentation.json ./shiro-init \ -m empty.hsmm \ -s unaligned-segmentation.json \ -FT > flat.hsmm

All of your instruction doesn't seem to got error.
I just download the CMU Arctic speech database and change the some instruction path. for example 'cmu_us_bdl_arctic/orig/' to 'cmu_us_bdl_arctic/wav/'(i can't find the orig path in dataset. i guess the wav path directory from orig to wav)

At first time, i try to inference the below code(I use ./examples/cmu-arctic-all-speakers.hsmm model instead trained-model.hsmm). but it also can't load the model. so i decide to train the another model through your training instructions.
./shiro-align \ -m trained-model.hsmm \ -s unaligned.json \ -g > initial-alignment.json ./shiro-align \ -m trained-model.hsmm \ -s initial-alignment.json \ -p 10 -d 50 > refined-alignment.json

from shiro.

Hello. How did you create flat.hsmm? Would you share the file for an inspection?

Sorry for my mistake to run the your instrunment.

I replace the path from "/cmu_us_bdl_arctic/orig/" to "/cmu_us_bdl_arctic/wav/".
But, the right way to modify is changing to "/cmu_us_bdl_arctic/wav"(please remove '/').
Finally, like the below figure, the training is successful!

All of top to end are work very well.
nice code!
Thanks a lot!

from shiro.