This brief presents an energy-efficient architecture to extract mel-frequency cepstrum coefficients (MFCCs) for realtime speech recognition systems. Based on the algorithmic property of MFCC feature extraction, the architecture is designed
with floating-point arithmetic.