<p>Language processing is essential for understanding the words spoken in one's native language. Tamil is rich in heritage and is considered one of the ancient languages. The grammar in the Tamil language is more complex and requires a deeper understanding and learning process. The sounds of a Tamil vowel differ in the pronunciation of each Tamil word containing vowels. In this work on Tamil vowel/word recognition, 1D and 2D features are extracted from spoken Tamil vowels/words, and templates are generated for each vowel/word using modelling techniques. Eighty per cent of the features are used for creating templates. Twenty per cent of the features are applied to the templates, and based on the matching between test features and templates, the test vector/matrix is classified as one of the Tamil vowels. Gammatone energy (GFE) features are derived from spoken utterances, and filter banks are calibrated at different scales. A random forest (RF) classifier is used to classify the Tamil vowel/word. 2D spectrogram features are derived from spoken utterances and given to the deep learning techniques to create convolutional neural network (CNN) templates. The system's performance is analysed using accuracy as a metric. The phase shift compensation is used for speech enhancement to remove the artefacts, and the system shows improvement in accuracy compared to the system without a speech enhancement mechanism.</p>