<p>This paper presents a new perspective to Malayalam speech recognition by integrating Multifractal Analysis using Continuous Wavelet Transform (CWT) and structure functions. Traditional speech recognition systems primarily rely on MFCC for feature extraction, which may overlook complex, non-linear characteristics. To address this limitation, a multifractal analysis framework that applies CWT to extract additional features reflecting the signal’s non-linearity and self-similarity at various scales is used in the proposed method. For the experiment, words have been selected from the LDC - IL Malayalam Raw Speech Corpus. Audio signals are initially processed using MFCC. Subsequently, Multifractal Analysis is employed to enhance these features. The Continuous Wavelet Transform (CWT)-based structure-function multifractal technique outperforms conventional traditional Multifractal Detrended Fluctuation Analysis (MF-DFA), Wavelet Leader Multifractal Analysis (WLMA) and Wavelet Transform Modulus Maxima (WTMM) methods. Experimental evaluation is conducted using a CNN classifier to determine the discriminative power of the extracted features. Experimental results indicate that MFCC combined with the proposed multifractal method achieved the highest accuracy of 95.19%, outperforming MF-DFA + MFCC (90.24%), WTMM + MFCC (94.45%), and WLMA + MFCC (93.57%). The standalone proposed method without MFCC yielded an accuracy of 56.51%, highlighting the complementary strength of MFCC features.This study advances speech processing by introducing a novel multifractal analysis method and integrating it with traditional feature extraction techniques, showcasing its effectiveness.</p>