Distributed%20Multimodal%20Synchronization%20Protocol%20(DMSP)

1
Distributed Multimodal Synchronization Protocol
(DMSP)

• Chris Cross
• IETF 65
• March 21, 2006

With Contribution from Gerald McCobb and Les
Wilson
2
DMSP Background

• Result of 4 year IBM and Motorola RD effort
• ID to IETF July 8, 2005 by IBM Motorola
• Reason for contribution
• A standard is needed to synchronize network based
  services implementing distributed modalities in
  multimodal applications
• Other protocols may have overlap but do not
  address all multimodal interaction requirements
• Other IETF IDs and RFCs
• Media Server Control Protocol (MSCP)
• LRDP The Lightweight Remote Display Protocol
  (Remote UI BoF)
• Media Resource Control Protocol Version 2
  (MRCPv2)
• Widex
• RFC 1056 Distributed Mail System for Personal
  Computers (also DMSP ?)

3
Why do you need a distributed system, i.e., a
Thin Client?
A thick client has speech recognition and
synthesis on the device. As resources available
on a device shrink or the application
requirements increase (larger application
grammars) then the performance of the system
becomes unacceptable. When that threshold is
reached then it is economically feasible to
distribute the speech over the network.
Grammar Size and Complexity G
R Resources memory and MIPS on the client
device G Size and Complexity of application
grammars R/G 1 Resources are adequate to
perform real time recognition and synthesis.
4
DMSP Architecture

• There are 4 DMSP building blocks

1. Modalities
2. Model-View-Controller (MVC) design pattern
3. View Independent Model
4. Event-based modality synchronization

5
DMSP Architecture Building Blocks

• Modalities are Views in the MVC Pattern
• GUI, Speech, Pen
• Individual browsers for each modality
• Compound browsers for multiple modalities

Compound Browser
6
DMSP Architecture Building Blocks

1. Model-View-Controller (MVC) design pattern

• Multimodal system can be modeled in terms of the
  MVC pattern
• Each modality can be decomposed and implemented
  in its own MVC pattern
• A modality can implement a view independent model
  and controller locally or use one in the network
  (e.g., an IM)

7
DMSP Architecture Building Blocks

• View Independent Model
• Enables a centralized model
• Modality interaction updates view and model
• Local event filters reflect important events to
  view independent model
• A modality listens to view independent model for
  only the events it cares about
• Compound clients, centralized control
  (Interaction Manager) as well as distributed
  modalities all enabled with a single protocol

8
DMSP Architecture Building Blocks

1. Event-based synchronization

• Compound Client All modalities rendered in
  client
• Interactions in one modality reflected in others
  thru event based changes to one or more model
• GUI DOM serves as View Independent model

9
DMSP Architecture Building Blocks

1. Event-based synchronization (CONTD)

• Distributed Modality A modality is handled in
  the infrastructure
• Requires the DMSP for distributing modality
• Event based synchronization via View Independent
  Model gives a modality independent distribution
  mechanism
• Enables multiple topographies
• Compound Client w/ Distributed Modality
• Interaction Manager

10
DMSP Design

• There are 4 abstract interfaces
• Command
• Response
• Event
• Signal
• Each interface defines a set of methods and
  related data structures exchanged between user
  agents
• Specified as a set of messages
• XML and Binary message encodings

11
DMSP Message Types

• Signals
• One-way asynchronous messages used to negotiate
  internal processing states
• Initialization (SIG_INIT)
• VXML Start (SIG_VXML_START)
• Close (SIG_CLOSE)

12
DMSP Message Types

• Command and control messages
• Add and remove event listener (CMD_ADD/REMOVE_EVT_
  LISTENER)
• Can dispatch (CMD_CAN_DISPATCH)
• Dispatch event (CMD_DISPATCH_EVT)
• Load URL (CMD_LOAD_URL)
• Load Source (CMD_LOAD_SRC)
• Get and Set Focus (CMD_GET/SET_FOCUS)
• Get and Set Fields (CMD_GET/SET_FIELDS)
• Cancel (CMD_CANCEL)
• Execute Form (CMD_EXEC_FORM)
• Get and Set Cookies (CMD_GET/SET_COOKIES)

13
DMSP Message Types

• Responses
• Response messages to commands
• OK (RESP_OK)
• Boolean (RESP_BOOL)
• String (RESP_STRING)
• Fields (RESP_FIELDS)
• Contains 1 or more Field data structures
• Error (RESP_ERROR)

14
DMSP Message Types

• Events
• Asynchronous notifications between user agents
  with a common data structure
• Events correlated with event listeners
• DOM events
• DOMActivate, DOMFocusIn, and DOMFocusOut
• HTML 4 events
• Click, Mouse, Key, submit, reset, etc
• Error and abort
• VXML Done (e.g., VoiceXML form complete)

15
DMSP Message Types

• Events (CONTD)
• Form Data
• One or more Field data structures (GUI or Voice)
• Recognition Results
• One or more Result data structures with raw
  utterance, score, and one or more Field data
  structures
• Recognition Results EX
• One or more Result EX data structures with raw
  utterance, score, grammar, and semantics
• Start and stop play back
• Play back of audio or TTS prompts has started or
  stopped
• Start and stop play back mark
• TTS encounters a mark in the play text
• Custom (i.e., application-defined)

16
DMSP Conclusion

• A protocol dedicated to distributed multimodal
  interaction
• Based on the Model-View-Controller design pattern
• Enables both Interaction Manager and Client based
  View Independent Model topographies
• Asynchronous signals and events
• Command-response messages
• Can be generalized for other modalities besides
  GUI and Voice
• Supports application specific result protocols
  (e.g. EMMA) through extension TBD
• Interested in getting more participation