An audio device may use the audio detected at two opposite facing, front and rear omnidirectional microphones to determine the angular directional location of a user's voice while the device in speaker mode or audio command input mode. The angular directional location may be determined to be at front, side and rear locations of the device during the period of time by calculating an energy ratio of audio signals output by the front and rear microphones during the period. Comparing the ratio to experimental data for sound received from different directions around the device may provide the location of the user's voice. Based on the determination, audio beamforming input settings may be adjusted for user voice beamforming. As a result, the device can perform better beamforming to combine the signals captured by the microphones and generate a single output that isolates the user's voice from background noise.