A system uses a voice energy detection (VED) circuit along with a capture buffer for improved performance and lower power dissipation. With a VED assigned to each microphone in the system, combing the output of more than one VED to improve the detection probability or reduce the false alarm rate improves the post detection signal-to-noise ratio. One or more VED circuits are dynamically selected based on background energy and detection performance. Further, a digital VED algorithm dynamically changes the oversampling ratio (OSR) values and passband bandwidth as a function of the loudness of the background noise and desired signal levels. If the desired signal or noise is strong, then the OSR is reduced to save power. If desired speech is detected, then the OSR value increases to get the target SNR for the remaining audio processing needs.