Attempting to deploy high-precision speech recognition algorithms on mobile and embedded devices without doing a thorough evaluation of their computational resource requirements is impractical. This study utilizes techniques to minimize the computational resources required for an optimized voice recognition system implemented on mobile devices. A cutting-edge voice recognition technique, known as Dynamic Multi-Layer Perceptron (DMLP), has been created. This method is capable of functioning in real-time on a contemporary mobile device. When evaluated on the same dataset, traditional hidden Markov models (HMM) performed somewhat better than our method, but this came at the cost of significantly longer processing time. The provided Dynamic Multi-layer Perceptron (DMLP) achieves a remarkable accuracy of 96.94% and outperforms other approaches in terms of speed.