Authors: S.Akhila, Ch.sathvika
Abstract: Due to its extensive usage in intelligent virtual assistants, healthcare systems, customer analytics, and human-computer interface, voice-based demographic prediction has become a significant field of study in speech processing. Developing personalised and context-aware services is made possible by accurately recognising demographic factors like age and gender from voice signals. For the purpose of determining an individual's age and gender from audio recordings, this research introduces a framework that combines deep learning with machine learning. To assess the benefits of sequential deep learning over traditional machine learning methods, we compare Logistic Regression's performance with that of an LSTM network for gender categorisation. Logistic Regression is used for multi-class classification after K-Means clustering is used to create synthetic age groups for age prediction. To accurately describe speaker-specific speech patterns, the suggested framework employs a number of spectral and acoustic parameters, such as statistical frequency descriptors, dominant frequency, interquartile range, spectral centroid, and mean fundamental frequency. In order to make the model more resilient and the predictions more accurate, we use thorough preprocessing, feature scaling, and feature selection. In terms of gender classification, the experimental evaluation shows that the LSTM model outperforms the baseline Logistic Regression model with a 98.5% accuracy rate, while the hybrid age prediction method successfully categorises synthetic age groups with high predictive performance. These results show that for intelligent applications in the real world, combining deep learning with traditional machine learning gives a scalable and dependable solution for voice-based demographic prediction.
