Automatic Speech Recognition (ASR) is a widely adopted technology with significant applications in home automation, enabling the accurate interpretation of spoken words and sentences. Integrating speech interfaces into Internet of Things (IoT) devices requires robust language models tailored to the target language, which in turn depend on comprehensive speech datasets. Although substantial progress has been achieved for well-resourced languages, ASR development for low-resource languages remains constrained by data scarcity. Addressing this challenge necessitates the creation of specialized datasets and frameworks. This study addresses this gap by developing a novel Telugu speech dataset and establishing a benchmark model utilizing Gaussian Hidden Markov Models (GHMM). Experimental results demonstrate that the proposed approach achieves 99 accuracy, indicating its effectiveness in IoT-based speech control systems. The performance of GHMM was compared with Support Vector Machines(SVMs) and Convolutional Neural Networks(CNNs), with GHMM outperforming both alternatives.