I'm training a world model and trying to use the generated acoustic model in sphinx 4. But it can not recognize anything. i'm actually experimenting with very small dataset, three utterances for five words each. while training using the runall.pl script i found a lot of error message in the html.log file like
....
Normalization for iteration: 1
norm Log File
This step had 4387 ERROR messages and 0 WARNING messages. Please check the log file for details.
completed
Current Overall Likelihood Per Frame = 17.5881022615536
....
=========================
and in the log files:
...
INFO: ........\src\programs\norm\main.c(472): Normalizing var
ERROR: "........\src\libs\libmodinv\gauden.c", line 1700: var (mgau= 30, feat= 0, density=0, component=0) < 0
ERROR: "........\src\libs\libmodinv\gauden.c", line 1700: var (mgau= 30, feat= 0, density=0, component=1) < 0
ERROR: "........\src\libs\libmodinv\gauden.c", line 1700: var (mgau= 30, feat= 0, density=0, component=2) < 0
ERROR: "........\src\libs\libmodinv\gauden.c", line 1700: var (mgau= 30, feat= 0, density=0, component=5) < 0
....
for all the normalizations for iteration 1 there're some errors but not in step 2 or 3. can anyone plz help? what's the reason for this?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
The reason is misunderstanding of the basic issues. You shouldn't train context dependant models with 1000 senones for a 3 words for sure. First of all answer the following questions:
What is the language you are training for.
What is the size of the vocabulary you'll use. Is it still 3 words or more?
Upload your files as well, it would be helpful for us to see them.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I'm training a world model and trying to use the generated acoustic model in sphinx 4. But it can not recognize anything. i'm actually experimenting with very small dataset, three utterances for five words each. while training using the runall.pl script i found a lot of error message in the html.log file like
....
Normalization for iteration: 1
norm Log File
This step had 4387 ERROR messages and 0 WARNING messages. Please check the log file for details.
completed
Current Overall Likelihood Per Frame = 17.5881022615536
....
=========================
and in the log files:
...
INFO: ........\src\programs\norm\main.c(472): Normalizing var
ERROR: "........\src\libs\libmodinv\gauden.c", line 1700: var (mgau= 30, feat= 0, density=0, component=0) < 0
ERROR: "........\src\libs\libmodinv\gauden.c", line 1700: var (mgau= 30, feat= 0, density=0, component=1) < 0
ERROR: "........\src\libs\libmodinv\gauden.c", line 1700: var (mgau= 30, feat= 0, density=0, component=2) < 0
ERROR: "........\src\libs\libmodinv\gauden.c", line 1700: var (mgau= 30, feat= 0, density=0, component=5) < 0
....
for all the normalizations for iteration 1 there're some errors but not in step 2 or 3. can anyone plz help? what's the reason for this?
The reason is misunderstanding of the basic issues. You shouldn't train context dependant models with 1000 senones for a 3 words for sure. First of all answer the following questions:
Upload your files as well, it would be helpful for us to see them.