Any breach of privacy and security in a digital medium is a serious issue. Keyboard acoustic emanation one of many side channel attack. The sound from keystroke of physical keys input can be reconstructed to extract information without the awareness of the user from a system. In this paper such keyboard acoustic emanations achieved by the process of record sample, filter sample, peak detection and classification for basic keystroke recognition. Cluster keystrokes provide the initial guess of sequence, both language model and supervised classifier work repeatedly classify, spellcheck over the result then retrieve classifier with corrected labels. The improved method increases the accuracy of text recovery of supervised data. The acoustic data of keystrokes properties and recording limiting factors are explored.
