Click here for full text:
Energy Aware Distributed Speech Recognition for Wireless Mobile Devices
Delaney, Brian; Simunic, Tajana; Jayant, Nikil
Keyword(s): low-power; distributed speech recognition; wireless
Abstract: The use of a voice-user interface for mobile wireless devices has been an area of interest for some time. However, these devices are generally limited by computation, memory, and battery energy, so performing high quality speech recognition on an embedded device is a difficult challenge. In this paper, we investigate the energy consumption of distributed speech recognition (DSR) on the HP Labs SmartBadge IV embedded system and propose optimizations at both the application and network layers that reduce the overall energy budget for this application while still maintaining adequate quality of service for the end- user. We consider energy consumption in both computation and communication. We present software optimization techniques that reduce the energy consumption of the speech signal processing algorithm by 83%. In addition, we estimate the energy consumption of client-side automatic speech recognition without the use of the network. We present a range of results such that the upper bound may match the results of server-based DSR and the lower bound offers reduced functionality (i.e. smaller vocabulary and/or lower accuracy) but with decreased energy usage. In our analysis of DSR, we consider both 802.11b and Bluetooth wireless networks. Given the relatively high bit rates these standards provide with respect to DSR traffic, we investigate the use of synchronous bursty transmission of the data to maximize the amount of time spent in a low-power or off state. The energy savings can be significant even with small, imperceptible delays. With 802.11b, we can reduce the energy consumption of the wireless interface by around 80% with modest application delays of just under half a second. We include the effects of a Rayleigh fading channel in our analysis and investigate the result of bit errors on both energy consumption and DSR accuracy. We have shown that DSR can reduce the required systemwide energy consumption for a speech recognition task by over 95% compared to a software based client-side speech recognition system. These savings include the software optimizations of the DSR front-end as well as the savings from the decreased duty cycle of the wireless interface. We have identified the lower bounds on channel SNR for the various network traffic types and have shown where it becomes advantageous or necessary to perform speech recognition on the embedded device.
Back to Index