Abstract. Speech recognition is a form of accessibility used to perform tasks with hands and eyes free, and this is advantageous regardless of the user type. Current APIs make implementation easy, but they have limitations because depend on Internet connectivity and, are often proprietary software. The proposed solution to resolve these limitations is the development of an off-line continuous speech recognition system. The best techniques selected in a systematic review were implemented using libraries. This work presents a comparative analysis of the evaluation metrics obtained for each library.