Wireless phrase-recognition PDA deployed to Afghanistan

Military Intelligence Professional Bulletin, Jan-March, 2003 by Lawrence Ricci, James Smith

A crucial component in supporting the commander's battlefield visualization requirement is communication which, for the purpose of this article, refers to the passing of information between an information source and an information user regardless of means. Conversation, whether face-to-face or by other means, is as effective today as it was in earlier times. Whether determining a threat's location, order of battle, tactics, strategies, or simply asking for directions, conversation has been, and will remain important in future military operations. A stumbling block, however, especially in locations that feature low-density languages, is the inability to translate a complete conversation, or at least the critical elements of that conversation. This article looks at a recently developed system that has already proven its utility and, as it improves, has great potential as a battlefield tool.

Need Realized During DESERT STORM

Soldiers' ability to communicate often limits effective employment of military forces in peacekeeping. During Operation DESERT STORM, mass surrenders of enemy troops overtaxed the coalition's ability to provide medical services. Battlefield doctors improvised a solution using laptop personal computers (PCs) with speech-recognition software, creating a list of frequently asked questions. When the user recognized a phrase, the PC played a WAV [sound] file of the Arabic equivalent. While the laptop had limited vocabulary and portability, the system proved useful. Later, a team headed by a former Navy SEAL (Sea-Air-Land special forces team member) realized that if they could package this function into a hand-held or belt-mounted computer, its utility would increase dramatically. They contacted the Defense Advanced Research Projects Agency (DARPA) about the project.

A number of companies developed the translator with about $1 million in grants from DARPA's Babylon program. The goal is to develop rapid, two-way, natural language, speech-translation interfaces and platforms for use in the field. Here the system will support force protection (FP), refugee processing, and medical triage. Babylon focuses on overcoming many technical and engineering challenges limiting current multilingual translation technology to facilitate future full-domain, unimpeded dialog translation in multiple environments. The Babylon program will focus on low-population, high-terrorist-risk languages that no commercial enterprise will support. The Babylon seedling one-way project, Rapid Multilingual Support (RMS), deployed to Afghanistan in spring 2002. DARPA has selected Mandarin and Arabic based on immediate and intermediate needs.

Context-Specific Speech-Recognition Software

Context-specific, speaker-independent software is in common use today for 1-800 telephone directory assistance, flight reservations, and other uses. Those applications, however, require server-sized computers. With the goal of bringing speech recognition into mobile embedded devices, the developers contracted to integrate the Phraselator[TM].

Ruggedized Low Power Consumption

This unit had to be ready for "instant on" phrase translation (e.g., "Stop or I'll shoot') while preserving full-day battery life. The development team started with a 32-bit RISC (reduced instruction set computer) CPU (central processing unit) capable of performing workstation-like speech analysis on little electric power.

After a certified process of uprating for temperature specifications, StrongARM[TM] provided the core of this system. The team carefully engineered the other components, including a high-fidelity, directional stereo-audio channel with both an inbound and outbound capability to maintain this low-power specification. The operating system offers both multimedia capability and comprehensive communication support for wireless applications. The developers of the system used this operating system to facilitate compatibility with pocket processor applications, and future versions will also be available with the upgraded operating system to encompass more robust network security and accomodate higher performance future CPUs.

Technical Challenges to Meet

A militarized personal digital assistant (PDA) has more requirements than the "traditional" speech recognition, including the ability to run in the toughest environmental conditions over extended temperature ranges in the harshest heat, rain, and snow. The touch-screen display must be readable in full sunlight and full darkness. They had to design a power system that could run for hours from multiple types of batteries (disposable, as well as rechargeable). These batteries would have to be able to accept a charge from various levels of applied current including 12-volts direct current (VDC) and 24-VDC vehicle power and 110or 220-volts alternating current (VAC).

The developers would have to use special care with the audio system design, allowing full-range frequency response from microphone through a compression/depression module (CODEC) and back out to the speaker. A goal for the speech-recognition capability was set at 98-percent accuracy in near-real time. Finally, the system had to be flexible and built to accept generally available third-party hardware and software.