<![CDATA[This study presents the design and implementation of a fully embedded offline voice control system for a Mecanum wheel robot integrating Whisper-based automatic speech recognition and ALBERT-based natural language understanding on a Raspberry Pi 4 platform. The proposed system supports parameterized motion commands with numerical value and unit extraction, enabling precise kinematic mapping without reliance on cloud services. The voice processing pipeline consists of audio acquisition, preprocessing, voice activity detection, speech transcription, intent classification, parameter extraction, kinematic transformation, and motor actuation. The system was trained on a custom dataset of 500 Indonesian navigation command samples spoken by five native speakers and evaluated on a separate test set of 200 commands. Experimental results demonstrate command recognition accuracy exceeding 95 percent and word error rates below 7.5 percent under moderate noise conditions. The system achieved an average end-to-end latency of 1.23 seconds. Motion execution errors remained below 5 percent within optimal parameter ranges, demonstrating sufficient precision for navigation tasks. Environmental robustness and reliability testing confirm stable performance in typical indoor environments. These results indicate that transformer-based speech and language models can be effectively deployed on resource-constrained embedded robotic platforms to enable practical real-time human–robot interaction. Specifically, the system addresses latency and privacy concerns associated with cloud-dependent solutions. The implementation demonstrates feasibility for educational and light industrial applications requiring offline capability.]]>
Copyrights © 2026
