Hi,
I would like to test the speech recognition with a set of custom features different from the MFCCs. I've trained the Italian acoustic model with this feature set by modifying the file "sphinx_train.cfg" (CFG_VECTOR_LENGTH = 20; CFG_FEATURE = '1s_c';).
I currently use PocketSphinx using MFCCs with good results for the Italian language but I've tried to do a recognition with my custom feature set, by changing the input parameters according to my vector length, and PocketSphinx not working. Particularly, by setting the length of my feature set to 13 (both in training that in recognition) recognition is successful; increasing the components of the set, PocketSphinx not working (locked in: "INFO: fsg_lextree.c(258)").
I would like to know if there are other changes that I have to do, or if PocketSphinx does not support custom features longer than the default value of 13.
Thank you for your time
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi,
I would like to test the speech recognition with a set of custom features different from the MFCCs. I've trained the Italian acoustic model with this feature set by modifying the file "sphinx_train.cfg" (CFG_VECTOR_LENGTH = 20; CFG_FEATURE = '1s_c';).
I currently use PocketSphinx using MFCCs with good results for the Italian language but I've tried to do a recognition with my custom feature set, by changing the input parameters according to my vector length, and PocketSphinx not working. Particularly, by setting the length of my feature set to 13 (both in training that in recognition) recognition is successful; increasing the components of the set, PocketSphinx not working (locked in: "INFO: fsg_lextree.c(258)").
I would like to know if there are other changes that I have to do, or if PocketSphinx does not support custom features longer than the default value of 13.
Thank you for your time
You can run pocketsphinx under debugger and press Ctrl+C to figure out the stacktrace where it locked.
Most likely it has no relation to feature length, it's more likely your custom features do not allow successful search.
You need to share the data files required to reproduce your problems to get help on this issue.