Are there any plans on supporting SRGS in PocketSphinx plugin with UniMRCP?
No, overall grammar-based approach is pretty broken. It does not match real human language capabilities, for that reason we consider examplar-based statistical models for the future probably instead of grammars. SRGS is low priority for us.
These days everyone else is moving to natural IVR with bigger vocabulary as well.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thanks for the response. Unfortunately it is very hard to change the way how Cisco works as the code is not made available to be changed. The only content type Cisco exposes is [text/uri-list] which is to load a grammar file from a URL or to specify the grammar itself as SRGS. Looks like my hands are pretty much tied to integrate CVP via MRCP unless you have any other recommendations. I'm not sure how JVoiceXML will tie into UniMRCP + PocketSphinx.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Another option I'm thinking is to create a Server side application which will process the audio file and return back text (like Google Speech Recognition service). Through the search of the forums, I was able to find a link to creating a servet but that post is very old (2008)
Hello Nickolay
Below is my current environment:
CentOS 7
UniMRCP v1.2
PocketSphinx/SphinxBase v0.8
MRCP v2
Cisco CVP
I am trying to integrate Sphinx with Cisco CVP environment using MRCPv2. I am getting the following errors from the application:
2015-10-26 14:22:33:389053 [INFO] Process DEFINE-GRAMMAR Request e84b8b1c7c1611e5@speechrecog [1]
2015-10-26 14:22:33:389302 [INFO] Dispatch Request DEFINE-GRAMMAR e84b8b1c7c1611e5@pocketsphinx
2015-10-26 14:22:33:389316 [WARN] Not Supported Content-Type [text/uri-list] e84b8b1c7c1611e5@pocketsphinx
2015-10-26 14:22:33:389328 [INFO] Process DEFINE-GRAMMAR Response e84b8b1c7c1611e5@speechrecog [1]
2015-10-26 14:22:33:389340 [INFO] Process DEFINE-GRAMMAR Request e84b8b1c7c1611e5@speechrecog [2]
2015-10-26 14:22:33:389516 [INFO] Dispatch Request DEFINE-GRAMMAR e84b8b1c7c1611e5@pocketsphinx
2015-10-26 14:22:33:389526 [WARN] Not Supported Content-Type [application/srgs+xml] e84b8b1c7c1611e5@pocketsphinx
2015-10-26 14:22:33:389536 [INFO] Process DEFINE-GRAMMAR Response e84b8b1c7c1611e5@speechrecog [2]
2015-10-26 14:22:33:389544 [INFO] Process RECOGNIZE Request e84b8b1c7c1611e5@speechrecog [3]
Are there any plans on supporting SRGS in PocketSphinx plugin with UniMRCP?
No, overall grammar-based approach is pretty broken. It does not match real human language capabilities, for that reason we consider examplar-based statistical models for the future probably instead of grammars. SRGS is low priority for us.
These days everyone else is moving to natural IVR with bigger vocabulary as well.
Also, version 0.8 is way less accurate than modern versions.
If you want to consider SRGS, it is somehow supported in JVoiceXML, which I never tried in practice.
Hi Nickolay
Thanks for the response. Unfortunately it is very hard to change the way how Cisco works as the code is not made available to be changed. The only content type Cisco exposes is [text/uri-list] which is to load a grammar file from a URL or to specify the grammar itself as SRGS. Looks like my hands are pretty much tied to integrate CVP via MRCP unless you have any other recommendations. I'm not sure how JVoiceXML will tie into UniMRCP + PocketSphinx.
JVoiceXML is another MRCP server, just like Unimrcp.
Hi Nickolay
MRCP support with JVoiceXML doesnt seem very robust as you pointed out, hence I will look for other options. Thanks.
Hello Nickolay
Another option I'm thinking is to create a Server side application which will process the audio file and return back text (like Google Speech Recognition service). Through the search of the forums, I was able to find a link to creating a servet but that post is very old (2008)
https://sourceforge.net/p/cmusphinx/discussion/sphinx4/thread/8b7f5499/?limit=250#b1de
Do you know if there are any new examples of creating such a service.
Again many thanks for all your help!