This work consists in developing a framework written in Java Micro Edition, for the recognition of abstract sounds and isolated words announcer dependent on mobile devices. Where was reached excellent recognition results, reaching – in the best case – an accuracy hits 94%. Through this framework, developers can use mobile applications and also work generally sound recognition in various scenarios. In making this proposal was discussed the MMAPI API to capture and execution of PCM audio encoding, already in phase of signal information extraction was used to extract the Mel-Cepstrais coefficients (MFCC) derived from the Fast Fourier Transform (FFT), and for the recognition was employed through comparison of Dinamic Time Warping (DTW). In this work were proposed two techniques of sound signal processing: the first called signal Inversion (IS), and the second corresponds to a low-cost algorithm for detecting extremes responsible for detection of beginning and end of an utterance. And also was proposed and integrated functionality that allows graphical representation of the audio spectrum on screens of mobile devices.