Spoken Command Recognition System for Hands-Free Door Control

This paper presents the development of a Spoken Command Recognition System (SCRS) that utilizes a lightweight convolutional neural network (CNN) for hands-free door access control. The system interprets voice commands, specifically "open" and "close", to activate a solenoid that locks or unlocks a door. An audio dataset was created by recording 1,000 open and close commands each and 1000 varied environmental noise. The dataset was augmented from 3,000 to 15,000 using a time mask, frequency mask, combined time and frequency mask, and noise addition. Mel-Frequency Cepstral Coefficients (MFCCs) were extracted from the audio samples; 80% of this was used for training and the remaining 20% for validation. The accuracy of the validation set is nearly 100%. The trained model was deployed on an ESP32-S3 microcontroller using TensorFlow Lite, and the system was packaged in a 155 mm × 155 mm × 56 mm air-vented metal. The deployed system achieved a recognition accuracy of 97.5% with a real-time response time of 3000ms for opening and closing door operations. The developed SCRS offers a low-power, real-time, and robust noise-tolerant solution for voice-activated door operations. It also offers an affordable solution with a cost of ₦71,500.