The scope of grammar support in Unimrcp SR plugins?

[Michael Levy]

Is there anything I can read that documents the scope of grammar support in the Unimrcp SR plugins? Or, can someone give me a brief explanation of the various supported grammar options?

I reviewed the usage manuals for the available plugins and for grammar support they seem to all support something like:

1. Built-in speech transcription grammar
2. Built-in and dynamic speech contexts
3. Built-in/embedded DTMF grammar
4. SRGS XML (limited support)

Is this correct:

1. Transcription - returns a transcription of the spoken phrase. 

2. speech contexts - defines a list of phrases in XML to use for the recognition and a tag string for the semantic result for the matching phrase

3. DTMF grammars seem limited to builtin:dtmf/digits. Do the plugins support VXML style minlength and maxlength querystring parameters?

4. SRGS XML - what is the scope of support for SRGS in these plug ins? The docs say "limited". What capabilities of SRGS are supported? Is there any SISR support?

I am trying to understand what are the supported options for semantic interpretation using the SR plugins?

How are people building speech applications and identifying the semantic meaning of the caller's words?

[Michael Levy]

I found this post which I think answers most of the above questions:

https://groups.google.com/forum/#!msg/unimrcp/13vVrW9cYTs/VRgSoHqDCgAJ

[Arsen Chaloyan]

Hi Michael,

This is a wide topic, and as you may see, not all the aspects are properly documented as of now.

Originally, there was no intention to support SRGS with new generation speech recognition engines, as transcription results are commonly passed to various NLP APIs for further processing. However, since the issue with SRGS support came up many times on different occasions, basic support has been in place for all the SR plugins for quite some time. They may not have reached parity yet.

There are various use cases: some need SRGS to have an actual grammar enforced, the others pass phrases used as a hint for speech transcription, not to mention, that meta data in SRGS XML also allows to pass additional vendor-specific parameters, and this turns out to be the most commonly supported option.

On Mon, Aug 26, 2019 at 11:46 AM Michael Levy <michae...@gmail.com> wrote:

I found this post which I think answers most of the above questions:

https://groups.google.com/forum/#!msg/unimrcp/13vVrW9cYTs/VRgSoHqDCgAJ

--
You received this message because you are subscribed to the Google Groups "UniMRCP" group.
To unsubscribe from this group and stop receiving emails from it, send an email to unimrcp+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/unimrcp/d5e7218e-4e82-41af-8f9a-969908553592%40googlegroups.com.

--

Arsen Chaloyan
Author of UniMRCP
http://www.unimrcp.org