i wanted to use the utility of vector quantization and after doing all the training stuff (deleted interpolation) is used it to get the quantized sub vectors. But there were certain errors which i encountered.
I hope that this problem can be resolved.
You have trained a semi-continous HMM in which only sphinx 3.0 supports it. It depends on what you want, you can ,
1, Train a semi-continuous HMM like what you did and use it with Sphinx3.0
2, Train a fully-continuous HMM, what you need to do is to re-train your model again by turning on the .cont. flag in sphinx_train.cfg. You can use this model in Sphinx 3.0, 3.x (x=4) and Sphinx 4.
I would also recommend you to take a look Dr. Rita Singh's document on Sphinx 3 training. It will give you a much better idea on what you were doing.
i wanted to use the utility of vector quantization and after doing all the training stuff (deleted interpolation) is used it to get the quantized sub vectors. But there were certain errors which i encountered.
I hope that this problem can be resolved.
Here is a sample run..
/usr/local/bin/gausubvq -mean /mnt/scsib/TrainingDBS/AN4/model_parameters/AN4.cd_semi_6000/means -mixw /mnt/scsib/TrainingDBS/AN4/model_parameters/AN4.cd_semi_6000/mixture_weights -subvq AN4.subvq.quant -svspec 0-12/13-25/26-38 -var /mnt/scsib/TrainingDBS/AN4/model_parameters/AN4.cd_semi_6000/variances
Argument 0: /usr/local/bin/gausubvq
Argument 1: -mean
Argument 2: /mnt/scsib/TrainingDBS/AN4/model_parameters/AN4.cd_semi_6000/means
Argument 3: -mixw
Argument 4: /mnt/scsib/TrainingDBS/AN4/model_parameters/AN4.cd_semi_6000/mixture_weights
Argument 5: -subvq
Argument 6: AN4.subvq.quant
Argument 7: -svspec
Argument 8: 0-12/13-25/26-38
Argument 9: -var
Argument 10: /mnt/scsib/TrainingDBS/AN4/model_parameters/AN4.cd_semi_6000/variances
INFO: cmd_ln.c(276): Parsing command line:
/usr/local/bin/gausubvq \ -mean /mnt/scsib/TrainingDBS/AN4/model_parameters/AN4.cd_semi_6000/means \ -mixw /mnt/scsib/TrainingDBS/AN4/model_parameters/AN4.cd_semi_6000/mixture_weights \ -subvq AN4.subvq.quant \ -svspec 0-12/13-25/26-38 \ -var /mnt/scsib/TrainingDBS/AN4/model_parameters/AN4.cd_semi_6000/variances
Configuration in effect:
[NAME] [DEFLT] [VALUE]
-eps 0.0001 1.000000e-04
-iter 100 100
-log3table 1.0003 1.000300e+00
-mean /mnt/scsib/TrainingDBS/AN4/model_parameters/AN4.cd_semi_6000/means
-mixw /mnt/scsib/TrainingDBS/AN4/model_parameters/AN4.cd_semi_6000/mixture_weights
-mixwfloor 0.0000001 1.000000e-07
-stdev 0 0
-subvq AN4.subvq.quant
-svqrows 4096 4096
-svspec 0-12/13-25/26-38
-var /mnt/scsib/TrainingDBS/AN4/model_parameters/AN4.cd_semi_6000/variances
-varfloor 0.0001 1.000000e-04
INFO: logs3.c(99): Initializing logbase: 1.000300e+00 (add table: -1834810029)
INFO: logs3.c(161): Log-Add table size = 29356
INFO: cont_mgau.c(95): Reading mixture gaussian file '/mnt/scsib/TrainingDBS/AN4/model_parameters/AN4.cd_semi_6000/means'
FATAL_ERROR: "cont_mgau.c", line 133: #Features streams(4) != 1 in continuous HMM
and before this run i used the parameters made after doing the deleted interpolation step and encountered the same error.
regards
BILAL AHMED
You have trained a semi-continous HMM in which only sphinx 3.0 supports it. It depends on what you want, you can ,
1, Train a semi-continuous HMM like what you did and use it with Sphinx3.0
2, Train a fully-continuous HMM, what you need to do is to re-train your model again by turning on the .cont. flag in sphinx_train.cfg. You can use this model in Sphinx 3.0, 3.x (x=4) and Sphinx 4.
I would also recommend you to take a look Dr. Rita Singh's document on Sphinx 3 training. It will give you a much better idea on what you were doing.
http://www.cs.cmu.edu/~rsingh/sphinxman/fr4.html
Regards,
Arthur