DeepVoice: An End-to-End Speaker Recognition System Leveraging Convolutional and Recurrent Neural Networks for Robust Voice Identification
S. Munavvar Hussain, Banala Saritha, B. Eswara Reddy, Chikile Srikar, B. Suchitra, G Purnachandrarao · 2025
In this work, we present DeepVoice, a comprehensive speaker recognition system that uses Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs) to improve the accuracy of voice-based identification systems. In contrast to conventional speaker recognition techniques that depend on segmented processing pipelines and manual feature engineering, DeepVoice uses automated feature extraction and temporal sequence modelling to expedite the recognition process. Because the system can handle both gender categorisation and speaker identification, it offers a flexible option for speech analysis jobs. Tests on a heterogeneous dataset with male and female speech samples show that DeepVoice outperforms traditional methods with 91.23% speaker identification accuracy and 98.56% gender classification accuracy with a validation loss of 0.21. These outcomes demonstrate how accurate and reliable the system is, which makes it a viable option for practical speaker identification applications.