The perceptual sound quality of desired audio signals (e.g., human voice) captured by an electronic device (e.g., cell phone) are improved by reducing ambient noise according to an algorithm that acts upon audio signals captured from a front and rear direction. More particularly, audio signals captured by two directional microphones pointing in opposite directions (e.g., a front microphone which receives audio signals from a forward direction and a rear microphone which receives audio signals from a rear direction) are classified and subsequently enhanced (e.g., unwanted signals are suppressed) according to a probability of their source (e.g., front, rear, or noise) thereby providing an improved perceptual sound recording than each microphone individually. The resultant signals provide decreased noise since the contribution of the front and rear microphones are taken into consideration and the signal from the more relevant (e.g., in the direction from which sound is coming) microphone is utilized.