MRCPv2-12 section 8.8 comments

View: New views
1 Messages — Rating Filter:   Alert me  

MRCPv2-12 section 8.8 comments

by David Copp :: Rate this Message:

Reply to Author | View Threaded | Show Only this Message

Hello, I am a member of Nuance's Platforms group and am writing an
MRCPv2 integration guide.

Section 8.8 "BARGE-IN-OCCURRED", middle paragraph, has a couple of minor
problems:


1. It refers to BARGE-IN-OCCURRED as an event:


   If a "SPEAK" request is active with kill-on-barge-in enabled, and the
   BARGE-IN-OCCURRED event is received, the synthesizer MUST immediately
   stop streaming out audio.


2. It seems to say that BARGE-IN-OCCURRED behaves rather like STOP:


   [The synthesizer] MUST also terminate any speech requests
   queued behind the current active one, irrespective of whether they
   have barge-in enabled or not. If a barge-in-able "SPEAK" request was
   playing and it was terminated, the response MUST contain the an [sic]
   active-request-list header listing the request-ids of all "SPEAK"
   requests that were terminated. The server generates no
   "SPEAK-COMPLETE" events for these requests.


This confuses the barge-in-able "SPEAK" request with pending "SPEAK"
requests. The reader is left thinking that ALL of the "terminated"
requests -- both the one that was barged-in upon and the ones behind it
-- do not get a "SPEAK-COMPLETE" event, which is not true.

A less confusing statement might be:


   [The synthesizer] MUST also terminate any "SPEAK" requests
   queued behind the current active one, irrespective of whether they
   have barge-in enabled or not. When queued "SPEAK" requests are
   terminated as a consequence of BARGE-IN-OCCURRED, the response MUST
contain an
   active-request-list header listing request-ids of the queued
   "SPEAK" requests that were terminated. The server generates no
   "SPEAK-COMPLETE" events for these requests.


Also note that in example 14.1, on page 173, the response to
BARGE-IN-OCCURRED lists the request ID of the preceding RECOGNIZE
request (543258) in the Active-Request-Id-List:


S->C: MRCP/2.0 72 543259 200 COMPLETE
      Channel-Identifier:32AECB23433801@speechsynth
      Active-Request-Id-List:543258
      Speech-Marker:timestamp=857206096314

It seems in this case the Active-Request-Id-List header must be removed from the response, since no queued "SPEAK" request
was terminated.


Cheers

David Copp
Platforms Group
Nuance Communications Inc.

_______________________________________________
Speechsc mailing list
Speechsc@...
https://www1.ietf.org/mailman/listinfo/speechsc
Supplemental web site:
<http://www.standardstrack.com/ietf/speechsc>
LightInTheBox - Buy quality products at wholesale price!