Chandigarh: AI Software Differentiates Human Voices from AI-Generated Ones
A groundbreaking artificial intelligence software developed by a forensic research team at Panjab University aims to differentiate between AI-generated and original human voices. Recently granted copyright by the Government of India’s Copyright Office, the software is spearheaded by Professor Kewal Krishan’s team in the anthropology department. Designed to combat the increasing prevalence of synthetic voices, this tool can be pivotal in forensic analysis of phone threats, criminal investigations, and security applications. The project, part of a Master’s dissertation by student Pratibha, involved contributions from several peers and used 100 audio samples—50 real and 50 AI-generated—to train a support vector machine model, achieving an impressive 80% accuracy. Completed over six months, Professor Krishan emphasized that this innovation could significantly aid forensic scientists in addressing issues like fraud and digital misinformation. Vice-chancellor Renu Vig highlighted the tool’s societal relevance and its potential in forensic applications.
