Navy transformations in operational planning, technology implementation, and personnel assignment point in the direction of changing the way warships are designed. Determining how ship maneuvering will be controlled occupies a central part in this transformation process. Replacement of humans with reliable technology and development of the littoral combat ship (LCS) bring to the fore an essential question regarding ship control: do ships need helmsmen? Speech recognition software is becoming more reliable and suitable for high-profile tasks. In 2003, the Naval Postgraduate School (NPS) conducted experiments at Marine Safety International in San Diego, California, to investigate the practicability of the technology for this purpose and identify key design elements for future testing.
Orders to the helm? Is there a helmsman who never requested that clarification-or a conning officer who never received that query? This standard phrase resolves miscommunication between the officer responsible for controlling shipboard maneuvers and the sailor who actually turns the ship's helm. It belongs to a set of long-established, standardized terms that marks the process of ship control in which every message or command must be clear and concise using a vocabulary common to both parties. The philosophy is that shiphandling should be as subconscious an act as speaking, and the "hardware" for converting speech to action is the helmsman. With the advent of speech recognition software (SRS), this manpower requirement may become obsolete. Before settling on a specific technology to integrate into future ship designs, the Navy should consider this ship control possibility.
The Navy recently initiated the design process for its proposed surface platform, the LCS. There is a growing interest in creation of a cockpit-like control station. In implementing technological innovations throughout the ship, planners should not overlook integrating voice-activated command control modules to provide reliable and efficient ship control. Specifically, there is an opportunity to replace human-in-the-loop conning with a voice-activated system (VAS). SRS transforms sound waves from the human voice into digital bits that get interpreted as commands and converted into mechanical or electrical signals. The resulting signals are relayed to the rudder and engine to adjust the angle and speed accordingly.
Surface naval forces are pursuing numerous labor-saving technologies and methodologies that yield reliable, effective shipboard operations. Acting Secretary of the Navy Hansford Johnson stated that one immediate goal is to "explore innovative manning initiatives such as the Optimum Manning Program, which relies on new technologies and creative leadership to reduce ship manning."1 Optimum manning program prototypes already exist in the USS Milius (DDG-69) and USS Mobile Bay (CG-53). On the Milius, optimum manning—part of the smart ship concept—operates with an "optimal crew size of just 232, almost 20% less crew than the usual complement for an Arleigh Burke (DDG-51)-class guided-missile destroyer." A goal in optimally manning current ships is to develop a functional requirement for the LCS and its expected crew size of 50.2
A helm watch is posted whenever a ship is under way, which requires a minimum of three helmsmen on eight-hour shifts. Moreover, minimally manning this watch station hazards crew fatigue, provides little redundancy, and leaves no room for training personnel for replacement, indicating a requirement for twice as many helmsmen and creating a demand on limited crew space. Technology can help reduce the manpower requirements by substituting for the helm, lee-helm, and- during special navigation details-helm safety observers. Deployment of a well-designed, technologically advanced conning system will greatly enhance the fighting power of LCS by keeping space available for more specialized crew.
Two methods may ease the manpower burden in ship control: directly employ the conning officer to manipulate ship controls in a cockpit-like structure or implement SRS to replace the helmsman. These alternatives require a careful investigation of development and implementation considerations, however, because implementation demands a reliable system that enables ready transition. Reliability does not refer only to system operational availability and proper translation of maneuvering thought into action. It also considers the users' sense of confidence and ability for command override.
In the cockpit option, equipping a console with every conceivable display, readout, and instrumentation can provide full awareness for shipboard maneuvering, and this is feasible technologically. The sole personnel requirement is to maintain a ship control watch, directly entering commands while at sea. Consolidating ship control functionality into a single station requires reengineering other navigation and monitoring capabilities so that the watchstander interacts with an integrated system. Ship control still will require actuation and there are a number of options available to a cockpit configuration, such as wheel and throttle actuators or keypad data entry. A third possibility is an Apache-helicopter-like headset that will maneuver in the direction the user looks. But this requires the conning officer to remain in a stationary position, which may decrease alertness. Yet another option is to integrate speech recognition software into the minimally manned pilothouse.
Over the past decade, SRS has become more common and in greater demand as improvements in accuracy and flexibility of commercial off-the-shelf (COTS) applications make this technology more user-friendly and feasible.3 Conning a ship is a prime example of human interaction where SRS could be instrumental. The conn's standard command, the helmsman's response and update of ship's status, and the conn's statement of understanding the maneuvering status is a clear, understandable process that transforms maneuvering a warship into an act as natural as speech itself. The VAS interactd with the ship system control segment of an integrated bridge to help reduce manning and preserve proven functions in an advanced form.4
Speech recognition is the process of converting an acoustic signal, captured by a microphone, into a set of words. The system translates the digital signal into the most likely of a restricted set of data candidates. In its final configuration, SRS must meet certain criteria for use on a U.S. warship:
* Accuracy rate equal to or greater than that of a human helmsman
* Ability to respond using verbal ship-handling vocabulary
* Use of standard conning commands
* Maneuverable support equipment
* Concise seamanship vocabulary
Note that the benefits of SRS technology also may be its weaknesses-just as the system converts verbal orders into actions, it may seek to convert spurious sounds and utterances as well. Consequently, high quality noise-cancellation capabilities are needed to reduce uncontrollable noise interference generated-for example, by the ship's engine or mechanical gear, environmental factors such as wind and rain, other watch standers, and bridge equipment.
Regardless of the conning system selected, it must be reliable, secure, trainable, mobile, and operable despite external disturbances. SRS is designed to use the same inputs as a human helmsman and must meet similar criteria. But a computer does not necessarily interpret commands delivered in the incorrect format; nor does it make adjustments for orders that do not match exactly what the conning officer intended-unlike what people can do. Conns must use the standard command set to match the system's predefined vocabulary. SRS technology is fallible (as is the current system). What experienced conn does not instinctively check indicators to verify that helm actions are as ordered? Functional requirements should focus on the operational setting of this system and include:
* Reliability. VAS must recognize and accurately relay commands, establishing a near-perfect user confidence level for the system reliability and accuracy.
* Multiuser. Ship control duties rotate among multiple users and must quickly and smoothly transition from one user to another. SRS capabilities must be impervious or adjustable to individual users' speech patterns, inflections, and accents.
* Security. Commanding officers require the ability to negate, interrupt, or override the commands of subordinates. Although verification or authentication guarantees that VAS software listens only to the authorized conn, it must be able to respond to an emergency order and disregard other voices.
* Mobility. A wireless headset will ensure orders may be delivered from bridge wings or wherever a shiphandling scenario leads the conn, much like the current scenario and unlike a cockpit's functionality.
* Resilience. Other crucial elements include the need for manual override, backup systems, and alternate power supplies for malfunctions and emergencies.
A new ship control system should reduce manpower demands and also fulfill functional requirements. Initial experimentation enabled some analysis of design and environmental factors influencing the operation of this critical system.
Before SRS is implemented, design engineers must determine whether it performs as well as the current conn-helmsman structure. Possible measures of effectiveness include rates of substitutions, insertions, and deletions by the software, as compared with inaction or incorrect activity by helmsmen. The NFS-Marine Safety International experiments were conducted in the spring of 2003 to determine significant design and environmental factors affecting COTS speech recognition software performances for conning applications. The study recorded errors and corresponding error rates.
The experiments considered the performance obtained with the Dragon Naturally Speaking (ver. 6.0) SRS product using five experienced individuals in a variety of conning scenarios, different ship types, and different simulator settings.5 The study was restricted to evaluation of the SRS performance; it did not analyze human error regarding specific helmsman performance. Each subject conducted between three and five trials. Scenarios included mooring, underway, and channel shiphandling in both Arleigh Burke-class destroyers and Oliver Hazard Perry (FFG-7)-class frigates. Results showed similar performances were obtained for successive trials for a given individual, setting, trial or scenario. Overall accuracy rates exceeded 90% throughout the experiment.
The study yielded important operational, design, and experimental insights. The bottom line is that SRS technology has the potential to change how Navy warships are driven in the future, but it requires full accounting of operational factors that would affect performance at sea. Although initial findings may appear to be mixed, they point in a direction that shows the potential of COTS SRS and identify issues to consider in further investigations. Initial findings include:
* Along with differences in individual style and manner, behavior such as coughing, bumping, and nose rubbing caused SRS to insert incorrect words. Different background noise levels and proximity of the wireless microphone to the SRS processor also had a significant effect on system performance. Good engineering design will have to account for these issues because these experimental factors-and some low-level interaction among them-accounted for more than 80% of the variation in SRS performance.
* Wireless systems are power intensive, requiring frequent battery changes. An indicator letting the user know of its current status is essential.
* Using the on-off switch incurs a delay from the time the microphone is turned on until it receives the signal, causing a lack of recognition. One option is to incorporate a capability for the microphone to go to a sleep or standby mode when a key word is spoken to disengage. For example, use the on keyword-such as "helmsman"—to activate and "very well" to deactivate SRS listening.
All Ahead Full
The Navy's vision to reduce future ship size and manning requirements indicates the need for increased reliance on technology to perform the functions currently performed by sailors and to support safe and sound ship handling. SRS has potential, and follow-on studies are under way to further investigate its potential for conning applications.
The experiments by NPS and Marine Safety International proved that voice-activated conning has great potential. At the same time, much exhaustive testing and related work remains to be done.
1 Statement of Honorable Hansford T. Johnson, Secretary of the Navy (Acting) Before the House Armed Service Committee, 23 February 2003. See www.chinfo.navy.mil/navpalib/people/secnav/johnsonht/testimony/johnson-hasc030226.txt,
2 Lt(j.g.) R. Flanders, USN, "First Optimally Manned Ship Aces Final Evaluation Problem," Navy Newsstand, 10 May 2002. (see www.news.navy.mil.); Littoral Combat Ship Flight 0: Preliminary Design Interim Requirements Document, Serial Number N763F-S03-026, 10 February 2003.
3 Janet Ray-Duprae, "Let's Talk," U.S. News and World Report, 12 May 2003, pp. 58-59.
4 Dorothy J. Tamez, "Using Commercial-Off-The-Shelf Speech Recognition Software for Conning U.S. Warships," master's thesis, Naval Postgraduate School, Monterey, California, June 2003.
5 D. Newman, Dragon Naturally Speaking (Software), (ver. 6), User's Guide, Scansoft Inc., 2001.
Dr. Fargues is an associate professor in the Electrical and Computer Engineering Department at the Naval Postgraduate School (NPS). Lieutenant Commander Gottfried, a surface warfare officer, serves as a military instructor in the Operations Research Department at NPS. Lieutenant Tamez is an information professional on the U.S. Pacific Command staff.