1-) Is there any negative effect of increasing S3_MAX_FRAMES in s3types.h
and
feat.h to decode large .wav files ?
Yes, it requires more memory and can even go out of memory. Whole decoding
history will
expand as a tree.
For that reason decoder architecture often includes endpointer, the
component that chunks the incoming continuous audio on small utterances.
Then each utterance is decoded separately. This endpointer could be
either embedded in decoder or can be standalone preprocessor as cont_ad
in sphinxbase. Sphinx 3 also has continuous tools like
sphinx3_continuous.
2-) Is there any tutorial how to use sphinx3_continuous ?
No, there is no such tutorial. We don't recommend you to use
sphinx3_continuous. We recommend our users to use pocketsphinx instead.
It provides consistent API for all required functionality including
n-best lists.
3-) How to get n-best hypothesis from large lattice files by using
sphinx3_astar
? (How to increase sphinx3_astar limits?)
Sorry, I'm not sure what limits are you talking about.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
3-) How to get n-best hypothesis from large lattice files by using
sphinx3_astar
? (How to increase sphinx3_astar limits?)
Sorry, I'm not sure what limits are you talking about.
I decoded a wav file which is length of 2.03 min. Than sphinx generated
related lattice file which has 15163 nodes in it.
When I run sphinx3_astar on this lattice file to get 10 best hypothesis, it
gives the following error:
INFO: dag.c(1158): Reading DAG file: ../lattice/test.lat.gz
INFO: dag.c(1314): dag->nfrm+1, 12334
INFO: main_astar.c(418): 12333 frames, 10876 nodes, 322494 edges, 0 bypass
INFO: astar.c(676): Writing N-Best list to ../nbest/test.nbest.gz
ERROR: "astar.c", line 607: Max PPATH limit (1000000) exceeded
ERROR: "astar.c", line 710: test: A* search failed
Should I increase this PPATH limit of sphinx3_astar or try something else like
pruning (beam or absolute).
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi,
I have to decode large .wav files (bigger than 2 min.) at one step and get
n-best hypothesis.
I have 3 questions about these.
1-) Is there any negative effect of increasing S3_MAX_FRAMES in s3types.h and
feat.h to decode large .wav files ?
2-) Is there any tutorial how to use sphinx3_continuous ?
3-) How to get n-best hypothesis from large lattice files by using
sphinx3_astar ? (How to increase sphinx3_astar limits?)
Yes, it requires more memory and can even go out of memory. Whole decoding
history will
expand as a tree.
For that reason decoder architecture often includes endpointer, the
component that chunks the incoming continuous audio on small utterances.
Then each utterance is decoded separately. This endpointer could be
either embedded in decoder or can be standalone preprocessor as cont_ad
in sphinxbase. Sphinx 3 also has continuous tools like
sphinx3_continuous.
No, there is no such tutorial. We don't recommend you to use
sphinx3_continuous. We recommend our users to use pocketsphinx instead.
It provides consistent API for all required functionality including
n-best lists.
3-) How to get n-best hypothesis from large lattice files by using
sphinx3_astar
? (How to increase sphinx3_astar limits?)
Sorry, I'm not sure what limits are you talking about.
I decoded a wav file which is length of 2.03 min. Than sphinx generated
related lattice file which has 15163 nodes in it.
When I run sphinx3_astar on this lattice file to get 10 best hypothesis, it
gives the following error:
INFO: dag.c(1158): Reading DAG file: ../lattice/test.lat.gz
INFO: dag.c(1314): dag->nfrm+1, 12334
INFO: main_astar.c(418): 12333 frames, 10876 nodes, 322494 edges, 0 bypass
INFO: astar.c(676): Writing N-Best list to ../nbest/test.nbest.gz
ERROR: "astar.c", line 607: Max PPATH limit (1000000) exceeded
ERROR: "astar.c", line 710: test: A* search failed
Should I increase this PPATH limit of sphinx3_astar or try something else like
pruning (beam or absolute).
Yes, it's maxppath option. You can decrease beams as well.