Pioneering Minimalist Speech Analysis Through Optimized Spectral Features Machine Learning Models

In the contemporary landscape, speech processing stands as a cornerstone across diverse applications, with gender identification emerging as a pivotal task in numerous domains, including law enforcement for swift resolution of criminal cases. The discernment of gender from speech recordings, especially in languages such as Telugu, holds paramount importance given its widespread usage and cultural significance in South India. This paper addresses the imperative need for robust gender identification methodologies tailored specifically for Telugu speech. Leveraging the rich heritage of Telugu as one of the most historically significant and widely spoken Dravidian languages, our study delves into the application of sophisticated statistical models, namely Gaussian Mixture Model (GMM) and Hidden Markov Model (HMM), to unravel the gender nuances embedded within Telugu speech patterns. To facilitate this identification task, we meticulously curated a bespoke dataset, drawing from diverse sources to ensure representativeness and reliability. Our methodology encompasses an array of advanced feature extraction techniques, including Mel Frequency Cepstral Coefficient (MFCC), augmented with first and second-order differentials, and Zero Crossing Rate (ZCR). Moreover, to streamline the analysis and optimize accuracy, we employ Principal Component Analysis (PCA) for feature reduction, thus enhancing computational efficiency without compromising on discernment accuracy. The efficacy of our approach is substantiated through rigorous evaluation and comparison analyses, wherein we meticulously assess the performance of both original and reduced feature sets. Our findings underscore the effectiveness of our methodology in achieving notable accuracy levels, thereby underscoring its potential for real-world applications in gender identification from Telugu speech data.