I am currently trying to comprehend the python wrapper for pocketsphinx and sphinxbase while reranking the output the following way:
1) I recognize input wav file and get best 10 candidates.
2) Then using the candidates I create a secondary test, where recognized phrases are 'quasiwords', which I transcribe and add to dictionary as a whole.
3) Using the 'vocabulary' I create a jsgf grammar with only transition such as: public <phrase> = hello-world | another-phrase | i-say-words
4) I try to recognize with this data but get an warning
WARN: "fsg_search.c", line 1508: Failed to find the end node
which follows the following info block
INFO: fsg_search.c(859): 0 frames, 0 HMMs (0/fr), 0 senones (0/fr), 1 history entries (0/fr)
In other words, for some reason, the language HMM was not built. And when I tried to recognize with the same data from console everything worked.
Can you please look into my problem? Thanks in advance!
PS: Will provide the code and data if needed.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
While preparing to provide the data I solved my problem :D It turned out that the when I was starting to decode for the second time I didn't open the stream for the recognition for the second time and referenced the same variable as before, since I didn't see the difference. But I find the whole thing quite strange.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hello!
I am currently trying to comprehend the python wrapper for pocketsphinx and sphinxbase while reranking the output the following way:
1) I recognize input wav file and get best 10 candidates.
2) Then using the candidates I create a secondary test, where recognized phrases are 'quasiwords', which I transcribe and add to dictionary as a whole.
3) Using the 'vocabulary' I create a jsgf grammar with only transition such as: public <phrase> = hello-world | another-phrase | i-say-words
4) I try to recognize with this data but get an warning
WARN: "fsg_search.c", line 1508: Failed to find the end node
which follows the following info block
INFO: fsg_search.c(859): 0 frames, 0 HMMs (0/fr), 0 senones (0/fr), 1 history entries (0/fr)
In other words, for some reason, the language HMM was not built. And when I tried to recognize with the same data from console everything worked.
Can you please look into my problem? Thanks in advance!
PS: Will provide the code and data if needed.
You need to provide the code and data
While preparing to provide the data I solved my problem :D It turned out that the when I was starting to decode for the second time I didn't open the stream for the recognition for the second time and referenced the same variable as before, since I didn't see the difference. But I find the whole thing quite strange.
Thank you for the, how it is put in russian, "оперативный" answer :)
I looked up some theoretical info on this case and it became clear now.