Exploiting acoustic emanations from electronic as well as mechanical devices as a means for side channel attack has recently emerged as a topic of security concern. In this paper, we present an attack methodology that can be used to extract the text typed by a user from the sound recorded by the built-in microphones of a mobile phone. We use signal processing techniques to initially extract a likely set of characters per tap on the touchscreen and then use natural language processing algorithms to find the most probable words and sentences that can be constructed from a given tap sequence. We also discuss the causes that result in this vulnerability and briefly present some countermeasures.