MRCPv2

1
MRCPv2

• Sarvi Shanmugham
• Cisco Systems Inc.

2
Status

• At version 00.
• Edits from the last meeting added.
• Awaiting addition of SI/SV functionality.
• A proposal for SI, SV, Enrollment, Hotword
  Recording is out as draft currently for MRCPv1
  support.
• Soon to be integrated into MRCPv2.

3
Open Issues

• Proxy support.
• Need to add call flows describing how MRCP
  proxies would work.
• Need to allow MRCP server/proxy to do a SIP
  re-INVITE and redirect the media as necessary.
• Starting/Stopping media from the client to the
  server for a RECOGNIZE method.
• Recording support
• Need support for a separate Recording resource.
• Does this recording resource relate to Recognizer
  capability to record utterances.
• Does this recording resource relate to Speaker
  Verification engines capability to record/buffer
  utterances.
• Does the utterance recording capability of
  Recognizers relate to Speaker verification needs
  to record utterances or buffer speech.

4
Open Issues

• Resource Types or Profiles
• Need to support separate resource types for
  things like DTMF recognizer, Audio Player,
  Poorman TTS.
• These may not require completely different
  resource state machines or methods and can be
  addressed by the same methods as Recognizer and
  Synthesizer resources. We could address them with
  resource sub-types or profiles. Any suggestions.
• NLSML Vs EMMA
• Proposal to add EMMA as a SHOULD have for MRCPv2.
• We would continue to have support NLSML for
  backward compatibility.

5
Open Issues

• Multiple media sessions under a single MRCP
  session.
• Do we need support for this.
• If there is a need for a separate media session
  for Recognition and SI and SV and Recording. Why
  have it under a single MRCP session? Would it be
  a reason to have a separate MRCP session?
• Do we need to support multiple active SPEAK
  requests with capability to switch between them
  by pausing and resuming the different speak
  request.
• Do we need to rename START-OF-SPEECH event to see
  that it applies for DTMF as well as speech.
  Should it be START-OF-INPUT?

6
Open Issues

• Content Management
• If we add support for Recorders, we may need
  mechanisms to fetch the recorded audio.
• Can we do this content management using HTTP
  instead of adding new methods into MRCPv2?
• Do we need to add support for Grammar management
  at the server level?
• Do we need the capability to delete grammars
  within a session?
• Do we need to support the management Speech
  content?
• Is there a need for Intermediate Recognition
  Results?
• Cookies?