| As part of the Future Warrior Technology Integration
(FWTI) program, Exponent completed the concept and development
phases of System Voice Control (SVC), a technology under investigation
at the U.S.
Army Natick Soldier Systems Center. The goal of SVC
was to integrate a viable Automatic Speech Recognition (ASR)
subsystem into a computer system that could enable the dismounted
soldier hands-free and, to some extent, eyes-free interaction
with the computer, radio, and other electronic subsystems.
During the SVC concept phase, Exponent performed a number
of tasks, including:
- Defined user and system requirements through a rigorous
systems engineering approach
- Developed a verification and validation test protocol
- Conducted a trade-off study of component products and
technologies
- Conducted a study of human factors issues
- Identified functional and interface requirements to facilitate
integration into the Land Warrior system
- Conducted a survey of current technology that included
research papers, government documents, and interviews with
Subject Matter Experts
- Developed use cases based on soldier feedback
- Identified feasible solutions through interactions with
potential vendors.
During the SVC development phase, Exponent defined domains
of variation over which ASR performance should be robust for
Objective Force Warrior systems, and quantified variations
in speech signal and noise and their influence on recognition
performance.
Using an iterative approach to building a functional, speaker-independent
demonstration system, Exponent provided the Government with
a flexible demonstration system and allowed the customer and
user representatives to have direct impact on the system design
and functionality.
Automatic Speech Recognition (ASR) Databases
In a related project, Exponent designed a process for the
creation of speech and complex noise databases for dismounted-soldier-worn
Automatic Speech Recognition (ASR) systems. The process involves
systematic collection of battlefield acoustic data (as it
occurs in carefully chosen training exercises), refinement
of the data into standardized databases useful for general
research, and validation of the databases via the demonstration
of improved ASR performance (when trained with the databases
versus when trained with noise-free data). Once the process
has been executed, these databases can be used to evaluate
competing ASR systems or to improve the battlefield performance
of a given ASR system.
top |
