Answers are grounded in this paper and cite their sources.
A dual-stage, ultra-low power (ULP) acoustic event detection (AED) system is presented. The proposed system consists of two stages, the always-on stage with ultra-low power serialized (ULPS) feature extraction, and the high performance stage based on long-term speech diversion (LTSD). The dual stage approach achieves ultra-low power consumption without significantly degrading system performance. A hardware-efficient ULP AED system architecture is discussed, and system performance evaluation is provided in details. The proposed system can achieve > 95% detection rate with 0.2% false alarm rate for voice activity detection (VAD) and 97% detection rate with 1% false alarm rate for AED while its estimated power consumption is in a nano-Watts range. By controlling the duty cycle ratio between two stages, the proposed system enables graceful tradeoff in system performance and power consumption.
No takes yet. Share an insight, caveat, or question.
Chen et al. (2016) studied this question.