I have a very small training set (less than 10 lines). I created it from samples of two sets I want to buy and combine; the point was to test that they would work together.
I am getting the error shown below. I think it may be caused by the small training set size. But, I'd like to be sure.
What issues could prevent the training sets from being combined? They are in the same format (sph) and were recorded at the same sample rate (16000 Hz). Is there anything else that could make them incompatible?
thanks!!
ERROR MESSAGE:
Baum welch starting for 1 Gaussian(s), iteration: 1 (1 of 1)
0% WARN: "...\sphinxtrain\src\libs\libio\model_def_io.c", line 436: Unable to open .../sphinx4/trainBox/test2/model_architecture/test2.1000.mdef for reading; FATAL_ERROR: "...\sphinxtrain\src\programs\bw\main.c", line 1054: initialization failed
Failed to start bw
Only 0 parts of 1 of Baum Welch were successfully completed
Parts 1 failed to run!
Training failed in iteration 1
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
> I have a very small training set (less than 10 lines).
That's not allowed
> WARN: "...\sphinxtrain\src\libs\libio\model_def_io.c", line 436: Unable to open .../sphinx4/trainBox/test2/model_architecture/test2.1000.mdef for reading; FATAL_ERROR: "...\sphinxtrain\src\programs\bw\main.c", line 1054: initialization failed
The source error was earlier in the logs. Probably it was a crash on previous stage due to the size of the db.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Did you try a tutorial? Each of 1000 (since you have 1000 tied states) samples you want to train now should have at least 50 representations in training data. Taking into account the average size of the unit about 0.05 ms it gives the about 1 hour data estimate.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I have a very small training set (less than 10 lines). I created it from samples of two sets I want to buy and combine; the point was to test that they would work together.
I am getting the error shown below. I think it may be caused by the small training set size. But, I'd like to be sure.
What issues could prevent the training sets from being combined? They are in the same format (sph) and were recorded at the same sample rate (16000 Hz). Is there anything else that could make them incompatible?
thanks!!
ERROR MESSAGE:
Baum welch starting for 1 Gaussian(s), iteration: 1 (1 of 1)
0% WARN: "...\sphinxtrain\src\libs\libio\model_def_io.c", line 436: Unable to open .../sphinx4/trainBox/test2/model_architecture/test2.1000.mdef for reading; FATAL_ERROR: "...\sphinxtrain\src\programs\bw\main.c", line 1054: initialization failed
Failed to start bw
Only 0 parts of 1 of Baum Welch were successfully completed
Parts 1 failed to run!
Training failed in iteration 1
> I have a very small training set (less than 10 lines).
That's not allowed
> WARN: "...\sphinxtrain\src\libs\libio\model_def_io.c", line 436: Unable to open .../sphinx4/trainBox/test2/model_architecture/test2.1000.mdef for reading; FATAL_ERROR: "...\sphinxtrain\src\programs\bw\main.c", line 1054: initialization failed
The source error was earlier in the logs. Probably it was a crash on previous stage due to the size of the db.
Thank you.
How small can a DB be to train successfully?
Did you try a tutorial? Each of 1000 (since you have 1000 tied states) samples you want to train now should have at least 50 representations in training data. Taking into account the average size of the unit about 0.05 ms it gives the about 1 hour data estimate.