Assessing Traditional and Deep Learning Voice Recognition Models for Reliable Control of Robots in Dynamic Settings

¹ Department of Automation & Robotics, KLE Technological University, Hubli, India
² Department of Computer Science, KLE Technological University, Hubli, India
³ Department of Electronics & Communication, KLE Technological University, Hubli, India

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Published online: 29 April 2026

Abstract

Net One of the biggest challenges in industrial robotics working in dynamic environments is reliably recognizing voice commands for human-robot interaction. This article presents a Hybrid Hidden Markov Model, Convolutional Neural Network (HMM-CNN) approach for voice- controlled collaborative robot operation. It was evaluated it against traditional Automatic Speech Recognition (ASR) frameworks, including Vosk and Kaldi. The HMM component captures phonetic patterns in voice data over time, while the CNN extracts key features from Mel-Frequency Cepstral Coefficient (MFCC) representations. The proposed system is trained on AI-generated voice samples that include “Pick” and “Place” commands spoken by 40 synthetic speakers. It has been validated it through real-time testing on the Omron TM5-700 collaborative robot and Webots simulation. The Vosk (87%) and Kaldi (89%) and HMM-CNN model achieves 93% command recognition accuracy, which is better than Vosk and Kaldi. Under different acoustic conditions it shows greater robustness. These demonstration shows how effective the hybrid architecture is for voice-controlled robot movement in dynamic industrial settings. It provides a basis for scalable, hands-free human-robot collaboration.