<p>Automatic speech recognition (ASR) systems are generally used to convert acoustic speech signals to their text transcriptions. This research proposes a model that combines a convolutional neural network with mel-frequency cepstral coefficients (MFCCs) for an isolated word-based ASR system. The Mel-Frequency scale identifies the most relevant features in speech samples. MFCC extracts speech features, which are fed as input to a convolutional neural network for the classification process. For the dataset, four speakers, out of which two male and two female speakers, were used to collect Dogri speech samples. Dogri is spoken primarily in the states of Jammu and Kashmir and some parts of Himachal Pradesh. The speakers selected for sample collection were natives of the Jammu region. The dataset consists of 20 words, for which five samples were collected from each speaker, making it a set of 20 samples per word, hence a total of 400 samples for 20 isolated words. These samples were recorded in a clean environment with no external noise. For the evaluation process, a confusion matrix was used to show how accurately the model identifies the spoken Dogri words. The confusion matrix clearly shows instances of classification and misclassifications. Parameters such as precision, recall, and F1-score are determined to demonstrate the model's efficiency. The developed model attained a precision of 0.8415, followed by the recall and F1-score of 0.8575 and 0.817, respectively. The good accuracy indicates the model’s efficiency in recognizing isolated words of the Dogri language. The model can be further extended for speech application in low-resourced languages. This research aims to create a speech recognition system with machine learning models such as convolutional neural networks and feature extraction techniques like MFCC to identify isolated words in the Dogri language.</p>