I'm trying to build a copy of sphinx 3.0 (slow decoder) from
the cvs distribution at archive_s3/s3. The code compiles ok,
but there seems to be only very limited documentation (e.g.
compared sphinx 2 or 3.3) on how to run it. No matter what
parameters, inputs, or models I use, the program s3decode
fails completely to recognize words, claiming
"search didn't end in </s>"
"NO RECOGNITION"
"Viterbi search failed"
This isn't just buggy recognition, it's no recognition. So I
suspect a configuration problem somewhere in my build,
e.g. a byte-order flag that's set wrong or the like.
I happen to have access to a working binary from a slightly
modified version of sphinx 3.0 used by the folks at HP
(formerly Compaq) Cambridge lab. That can recognize my
input files ok with the exact same acoustic/language models
and parameter settings.
So, I suspect there's something wrong with my build of sphinx
3.0, but that distribution doesn't seem to include any test
scripts and I'm at a loss about how to track down the source
of the problem.
I'm running Redhat 7.2 linux and taking input from a file
(i.e. not a microphone). sphinx 2 and 3.3 run ok for me.
Any suggestions would be greatly appreciated.
In particular, sphinx 3.3 comes with a test script that runs
the 3.3 engine on a test input file with the supplied acoustic
and language models. Can anyone supply me with a
similar script that runs the sphinx 3.0 engine on the
same example with the same models, so I can see what
parameter settings are required to make the 3.0 engine work
right on at least one example input?
????
Margaret
Margaret Fleck, HP labs
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I'm trying to build a copy of sphinx 3.0 (slow decoder) from
the cvs distribution at archive_s3/s3. The code compiles ok,
but there seems to be only very limited documentation (e.g.
compared sphinx 2 or 3.3) on how to run it. No matter what
parameters, inputs, or models I use, the program s3decode
fails completely to recognize words, claiming
"search didn't end in </s>"
"NO RECOGNITION"
"Viterbi search failed"
This isn't just buggy recognition, it's no recognition. So I
suspect a configuration problem somewhere in my build,
e.g. a byte-order flag that's set wrong or the like.
I happen to have access to a working binary from a slightly
modified version of sphinx 3.0 used by the folks at HP
(formerly Compaq) Cambridge lab. That can recognize my
input files ok with the exact same acoustic/language models
and parameter settings.
So, I suspect there's something wrong with my build of sphinx
3.0, but that distribution doesn't seem to include any test
scripts and I'm at a loss about how to track down the source
of the problem.
I'm running Redhat 7.2 linux and taking input from a file
(i.e. not a microphone). sphinx 2 and 3.3 run ok for me.
Any suggestions would be greatly appreciated.
In particular, sphinx 3.3 comes with a test script that runs
the 3.3 engine on a test input file with the supplied acoustic
and language models. Can anyone supply me with a
similar script that runs the sphinx 3.0 engine on the
same example with the same models, so I can see what
parameter settings are required to make the 3.0 engine work
right on at least one example input?
????
Margaret
Margaret Fleck, HP labs