CAE CAV HMC MMC OSD PAF
AI Workflows AI Modules

1       AI Workflows

1.1      Communicating Entities in Context

HMC-CEC is a PAAI that communicates with a human in a real environment and/or an HMC-CEC who may be rendered as a Persona in an M-Instance.

It composed of a set of the following collaborating PAAIs:

Audio-Visual Scene Description Describes the real scene as Audio-Visual Scene Descriptors
Audio-Visual Scene Integration and Description Extracts the components of the Communication Item (Personal Avatar) produced by another HMC-CEC PAAI.
Entity and Context Understanding Extracts the components of the Communication Item providing:
– Entity ID
– Text Object
– Meaning of Text Object
– Personal Status of Entity
– AV Scene Geometry
– Instance ID of Audio/Visual Object.
Entity Dialogue Processing Reacts to the six inputs above by producing:
– Its own ID
– A Text Object
– Meaning of its own Text Object
– Its own Personal Status.
Text-to-Text-Translation Translates its own Text
Personal Status Display Produces a Communication Item containing its reaction.
Audio-Visual Scene Rendering Renders its Communication Item for use in the real world.

Figure 12 – Reference Model of HMC-CEC

The following links analyse the AI Modules:

AV Scene Integration and Description

Entity and Context Understanding

2       AI Modules

2.2      Entity and Context Understanding

HMC-ECU is a PAAI composed of the following collaborating PAAIs:

Audio-Visual Scene Demultiplexing Provides the Descriptors of the Scene.
Automatic Speech Recognition Converts Speech Object into Text Object.
Visual Object Identification Provides the ID of a Visual Object with reference to a Taxonomy.
Audio Object Identification Provides the ID of a Visual Object with reference to a Taxonomy.
Natural Language Understanding Provides Refined Text and Meaning.
Personal Status Extraction Provides the Personal Status of the Entity.
Text-to-Text Translation Provides a translation of the Text Object.

Figure 13 – Reference Model of CUI-ECU

The implementation of HMC-ECU depends on the implementation of its components:

  • Speech Recognition – Neural Network
  • Visual and Audio Object ID Recognition – Neural Network
  • Natural Language Understanding enhanced by Scene Geometry
  • Personal Status Extraction – partly Neural Networks,

HMC-ECU performs Interpretation Level Operations.