Application Of Kalman Filter Algorithm to Reduce Noise In Door Lock Opening System With Voice Recognition

Abstract

Acoustic noise refers to unwanted sound disturbances occurring during data recording and collection. This interference often mixes with desired signals, creating challenges in accurately detecting intended voice commands. This study focuses on the implementation of the Kalman Filter Algorithm to mitigate acoustic noise in a voice recognition-based door lock system. The system utilizes an INMP441 microphone module interfaced with an ESP32 microcontroller, which transmits audio data to an edge server. The Kalman Filter Algorithm is deployed on the ESP32 to denoise audio signals from environmental interference. Keyword detection is performed using six comprehensive analysis algorithms, which examine syllable structure, energy patterns, consonant-vowel differentiation, temporal patterns, spectral analysis, and voice characteristics. Experiments were conducted under various noise intensity levels and pronunciation volume variations, comparing performance both with and without the Kalman Filter. The results indicate that the Kalman Filter improved system accuracy from 69.6% to 83.2%, while effectively mitigating false positives—instances where incorrect inputs are erroneously validated as correct. The system successfully recognized the keyword "buka" (open), achieving an average response time of 6.54 seconds to unlock the door following keyword confirmation by the edge server. However, high-intensity noise, such as that produced by grinding machinery, presents significant detection challenges. Recommendations for future research include enhancing microphone sensitivity and developing adaptive keyword detection methods to manage diverse acoustic environments and noise conditions more effectively.