This article presents an empirical approach for isolated word and phoneme-level speech recognition in the Azerbaijani language. The method integrates three complementary techniques: Dynamic Time Warping (DTW) and spectral analysis based on Fourier (FT) and wavelet (WT) transforms applied to continuous speech signals. Each technique captures distinct temporal and frequency-domain features, enabling a comprehensive representation of phonetic and lexical information. Recognition is performed using an algorithm that evaluates informative features and selects an appropriate similarity measure for one-dimensional signals, coordinating time-domain alignment with multiresolution spectral features. The framework is particularly suited for languages with limited annotated speech resources, supporting effective recognition at both phoneme and word levels. Experimental results demonstrate that the combined use of DTW with Fourier- and wavelet-based features improves recognition accuracy over individual methods. These findings confirm the feasibility of the proposed approach for Azerbaijani speech recognition and highlight its potential applicability to other low-resource languages and speech processing tasks.