This Paper outlines the creation, construction, and evaluation of a Real-Time Voice Recognition System using TinyML on an Arduino Nano 33 BLE, a hardware platform with limited resources. Tiny ML consists of machine learning models that can be deployed on the low -energy and resource- constrained embedded system. Real-time voice Recognition is used in virtual assistants like Siri, Google Assistant, Alexa, speech-to-text translation, vehicle Navigation, voice biometrics, home automation etc. The work presents the implementation of real-time voice recognition wherein a TinyML model is trained using the EdgeImpulse framework and deployed on Arduino Nano 33 BLE device having a built-in microphone. The model identifies the ON and STOP keywords pronounced by the user using Arduino Nano 33 BLE device's built-in microphone which then turns on a bulb connected to it via a relay. The accuracy of the system is found to be 97 %.