This paper discusses the application of the Hidden Markov Model (HMM) for isolated word speech recognition, specifically focusing on building a user-dependent system trained on a limited set of 105 variations of seven words recorded by a single speaker. The system employs algorithms such as Baum-Welch and Gaussian fitting to optimize performance, resulting in an efficiency increase from 15% to about 95%. Primary applications include voice-activated systems for individuals with impairments and security systems requiring voice recognition.