Optimized radix-2 FFT and Mel-filter bank in MFCC-based events sound recognition chip design for active smart warming care

The paper proposes a first sound chip design for security-sensitive event sounds recognition that extended the interaction of Orange warming care from human-to-human to environment-to-human perception. The proposed chip is fittingly embedded in smart sensors or appliances at home to surroundingly detect the event sounds, which can timely care the elderly or children who live alone thus actively call for assistance. In order to realize the chip in a high-accuracy performance, a small-size area and a low-power dissipation, the MFCC several sub-modules including, radix-2 FFT, Mel-filter bank etc are optimized for chip design to reach the required characteristics. In the simulation results, the proposed MFCC with k-NN framework performs the higher recognition accuracy than LPCC and MP features having k-NN classifier. For chip realization, the optimized MFCC sub-modules indeed improve the hardware resource utilization, where the chip is designed and simulated by verilog and synthesized by TSMC 90nm library.