Appendix E Wearable Speech-Operated Computer
Figure E-1 depicts the evolution of a speech-operated system implemented in software from a standard laptop computer to a belt-worn computer to a hand-held device. The effectiveness of the design is measured by a figure of merit defined by real time response divided by the volume multiplied by the weight and the power normalized to unity for the laptop. As described in this appendix, the effectiveness of the system was improved by almost three orders of magnitude by improving volume, weight, and power consumption by at least a factor of five.
Figure E-2 depicts two systems and shows the impact of power management techniques. Navigator 1 was a first generation design for a speech-operated system and was based on an Intel 80386, 25 MHz processor with 16 MBytes of main memory and an 85 MByte disk drive. The system was controlled by speech recognition software. Once the basic hardware configuration was selected, the energy usage of the system was reduced by more than a factor of two using the following techniques:
-
Operating system software. A profile of computer usage revealed that the system never took advantage of a low energy-consuming standby mode of operation. Further examination indicated that even when the system was idle the operating system was making more than 150 different types of calls to system utilities that consumed energy, not only in the processor but also in the peripheral devices, such as the hard disk drive by keeping it actively spinning. By replacing the functionality of the old idle loop with a halt instruction, the processor was able to enter standby mode and save more than a factor of two in energy consumption.
-
Hard disk drive. The disk drive was placed in standby mode when it was not required for loading programs.
-
Application profile. The speech-driven application was profiled to determine which system resources were consumed. A number of peripheral buses and input/output ports were not required. These digital signals were terminated by energy consuming resistors. Thus, substantial energy was saved by permanently disabling resources that were not used by the application. In addition, the power supply was redesigned to increase efficiency when delivering multiple voltages from the battery subsystem.
In the second generation system, called Navigator 2, a hardware system consisting of an Intel 80486 processor operating at 33 MHz with 12 MBytes of main storage and a 420 MByte disk drive was used. The system was designed for use during the inspection of sheet metal on aircraft. The following techniques cut energy consumption of the basic hardware system by a factor of four:
-
Processor clock frequency. The processor clock frequency was lowered to the minimum that could provide an adequate response for the application. Rather than seeking a software speech recognition solution that would have required at least 100 MIPS (as was done in Navigator 1), a separate dedicated speech application PCMCIA card, based upon a 13 Mhz DSP, allowed the selection of a more energy-efficient processor.
-
Hard disk drive. The software was modified to be core-resident to avoid paging from the hard drive. Thus the hard drive could be disabled except for program loading and database updates.
-
Speech. A separate "onset of speech" recognition circuit enabled the speech recognition PCMCIA card only when there was actual speech to be recognized. Otherwise the speech recognition card was disabled.
-
Display. The display was disabled when it was moved from in front of the eyes or after a period of inactivity. The "see-through" display allowed viewing of the user's environment without physically removing the display.
Power reductions made over the two generations of the Navigator speech-operated system are examples of what can be accomplished using standard PC hardware by moving from general-purpose operation to function-specific operation. Much greater improvements, by factors of 10 to 100, can be achieved by moving away from PC hardware platforms to dedicated embedded systems.