MRCP architecture with multiple servers?

[Erland Lewin-2]

I am trying to understand the MRCPv2 specification. Please correct me if
I've misunderstood something.

Refering to Figure 1: Architectural Diagram, I believe I understand it.

How would that architecture look if there were two simultaneous MRCPv2
Media Servers used by one client? I assume that this might be the case
if one would want to use speech processing resources from different
vendors, which do not plug into the same Media Resource Server
implementation (since the "Media Resource Management" layer is not
standardized).

If I understand correctly, one consequence of multiple Media Resource
Servers would be that one would require a component which mixes and
divides the RTP streams from the Media Source/Sink to each Media
Resource Server. How would this component fit into the MRCP framework?

Would a clarification of this situation merit mention in the specification?

Regards,

Erland Lewin
Voxi AB




_______________________________________________
Speechsc mailing list
Speechsc@...
https://www1.ietf.org/mailman/listinfo/speechsc
Supplemental web site:
<http://www.standardstrack.com/ietf/speechsc>

[David R Oran]

On Sep 6, 2007, at 4:38 AM, Erland Lewin wrote:

> I am trying to understand the MRCPv2 specification. Please correct  
> me if I've misunderstood something.
>
> Refering to Figure 1: Architectural Diagram, I believe I understand  
> it.
>
> How would that architecture look if there were two simultaneous  
> MRCPv2 Media Servers used by one client? I assume that this might  
> be the case if one would want to use speech processing resources  
> from different vendors, which do not plug into the same Media  
> Resource Server implementation (since the "Media Resource  
> Management" layer is not standardized).
>
> If I understand correctly, one consequence of multiple Media  
> Resource Servers would be that one would require a component which  
> mixes and divides the RTP streams from the Media Source/Sink to  
> each Media Resource Server. How would this component fit into the  
> MRCP framework?
>

(With individual contributor hat on and co-chair hat off).

It could one of a number of things:
a) if the media source is in the MRCP client, the client could simply  
produce two RTP streams
b) it could be an RTP mixer-translator, a defined architectural  
element of the AVT archiecture (not defined specifically by MRCP/
SPEECHSC).
c) the RTP stream could be multicast and the media server(s) simply  
join the multicast group.

> Would a clarification of this situation merit mention in the  
> specification?
>
It would add a lot of use cases. I'm not sure it's worth adding a lot  
of verbiage to cover such cases at this late date.

Note also, that for some uses of MRCP, like text-to-speech, the multi-
stream scenario doesn't arise even with multiple media servers,  
because in this case there is only one receiver (the client) and both  
media servers can send RTP to this client n a single RTP session.

Cheers, Dave.

> Regards,
>
> Erland Lewin
> Voxi AB
>
>
>
>
> _______________________________________________
> Speechsc mailing list
> Speechsc@...
> https://www1.ietf.org/mailman/listinfo/speechsc
> Supplemental web site:
> <http://www.standardstrack.com/ietf/speechsc>

_______________________________________________
Speechsc mailing list
Speechsc@...
https://www1.ietf.org/mailman/listinfo/speechsc
Supplemental web site:
<http://www.standardstrack.com/ietf/speechsc>