Speech Recognition Engines

Flat
sphinx 3 Forward-Backward error
User: chn
Date: 7/21/2008 9:34 pm
Views: 13080
Rating: 7

Phase 3: Forward-Backward

Baum welch starting for iteration: 1 (1 of 1)

bw Log File

completed

Normalization for iteration: 1

norm Log File

completed

Current Overall Likelihood Per Frame = 0.359086967147846

Baum welch starting for iteration: 2 (1 of 1)

bw Log File

This step had 2 ERROR messages and 1 WARNING messages. Please check the log file for details.

completed

Normalization for iteration: 2

norm Log File

WARNING: This step had 0 ERROR messages and 3 WARNING messages. Please check the log file for details.

completed

Current Overall Likelihood Per Frame = 3.29868117388543

Convergence Ratio = 8.18630158060643

Baum welch starting for iteration: 3 (1 of 1)

bw Log File

This step had 2 ERROR messages and 1 WARNING messages. Please check the log file for details.

completed

Normalization for iteration: 3

norm Log File

WARNING: This step had 0 ERROR messages and 3 WARNING messages. Please check the log file for details.

completed

Current Overall Likelihood Per Frame = 4.23156582492712

Convergence Ratio = 0.28280534003317

Baum welch starting for iteration: 4 (1 of 1)

bw Log File

This step had 2 ERROR messages and 1 WARNING messages. Please check the log file for details.

completed

Normalization for iteration: 4

norm Log File

WARNING: This step had 0 ERROR messages and 3 WARNING messages. Please check the log file for details.

completed

Current Overall Likelihood Per Frame = 4.34869005022033

Training completed after 4 iterations

 log file:

utt>   545               cen2-mjdr-b  199    0    36 11 ERROR: "backward.c", line 431: final state not reached
ERROR: "baum_welch.c", line 331: mjdr/cen2-mjdr-b ignored


50 Training Context dependent models

 


 INFO: main.c(474): Normalizing var
ERROR: "gauden.c", line 1700: var (mgau= 712, feat= 0, density=0, component=1) < 0
ERROR: "gauden.c", line 1700: var (mgau= 712, feat= 0, density=0, component=2) < 0

'''''''''more ERROR 

--- (Edited on 7/21/2008 9:34 pm [GMT-0500] by chn) ---

Re: sphinx 3 Forward-Backward error
User: nsh
Date: 7/22/2008 12:58 am
Views: 92
Rating: 6

Read this thread:

http://www.voxforge.org/home/forums/message-boards/speech-recognition-engines/doubt-about-sphinx3-installation

 

--- (Edited on 7/22/2008 12:59 am [GMT-0500] by nsh) ---

Re: sphinx 3 Forward-Backward error
User: chn
Date: 7/22/2008 2:37 am
Views: 101
Rating: 10

30 Training Context Dependent models

Every BW step had 2 ERROR messages and 1 WARNING messages.

And you said it can be ignored .  But

 50 Training Context dependent models

BW step had 2 ERROR messages and 1 WARNING messages;

But norm step:

Some of this steps cound have up to 2116 ERROR messages and 0 WARNING messages.

 

 

--- (Edited on 7/22/2008 2:37 am [GMT-0500] by chn) ---

Re: sphinx 3 Forward-Backward error
User: nsh
Date: 7/22/2008 3:10 am
Views: 105
Rating: 7
Ignore them as well. The reason of the problem is that you are trying to train 1000 senones on a small amount of data. Either use more training data or lower the amount of senones in config file.

--- (Edited on 7/22/2008 3:10 am [GMT-0500] by nsh) ---

Re: sphinx 3 Forward-Backward error
User: Visitor
Date: 7/22/2008 7:05 am
Views: 96
Rating: 8

Thanks for you help.

And I set senones=500,then I got SE:50.6%,WE: 14.9%

In the process of Training , the Errors were down to 184(the most)

--- (Edited on 7/22/2008 7:05 am [GMT-0500] by Visitor) ---

Re: sphinx 3 Forward-Backward error
User: Visitor
Date: 7/22/2008 7:22 am
Views: 112
Rating: 6

Thanks for you help.

And I set senones=500,then I got SE:50.6%,WE: 14.9%

In the process of Training , the Errors were down to 184(the most)

--- (Edited on 7/22/2008 7:22 am [GMT-0500] by Visitor) ---

Re: sphinx 3 Forward-Backward error
User: chn
Date: 7/22/2008 7:25 am
Views: 109
Rating: 9
Whether I can reduce ERROR down to zero or not?

--- (Edited on 7/22/2008 7:25 am [GMT-0500] by chn) ---

Re: sphinx 3 Forward-Backward error
User: nsh
Date: 7/22/2008 3:11 pm
Views: 97
Rating: 7

The core thing in ASR is that error always present. You can't be 100% sure you recognized correctly, you can only get some probability. That's why speech interfaces always use confidence and cross referencing to setup agreement between human and the system. Of course, you coudn't rely on recognition result.

For performance rate you can reach, everything depends on the size and the complexity of the vocabulary. Typical error rate for digits is 0.5%, for large vocabulary it's around  15-25%.

--- (Edited on 7/22/2008 3:11 pm [GMT-0500] by nsh) ---

Re: sphinx 3 Forward-Backward error
User: chn
Date: 7/22/2008 7:52 pm
Views: 497
Rating: 6
OK,thank you very much.

--- (Edited on 7/22/2008 7:52 pm [GMT-0500] by chn) ---

Re: sphinx 3 Forward-Backward error
User: Visitor
Date: 5/20/2010 1:37 am
Views: 109
Rating: 2

Can you tell me what mean of senones . and how can i set it

Thank

--- (Edited on 5/20/2010 1:37 am [GMT-0500] by Visitor) ---

PreviousNext