Chandigarh, Sept 23 — A nine-member research team from Panjab University’s department of anthropology has developed an artificial intelligence (AI) tool that can tell apart cloned voices from genuine human speech, offering new support for forensic and security investigations.
The team, led by master’s student Pratibha and supervised by forensic scientist Professor Kewal Krishan, trained the model using 100 audio samples — 50 natural voices and 50 computer-generated. The system, based on a Support Vector Machine (SVM) model, was designed to pick up subtle differences in tone, speech texture and individual vocal patterns. During testing, it achieved an accuracy rate of 80 percent.
The project formed part of Pratibha’s master’s dissertation and has now been granted copyright by the Government of India’s Copyright Office. According to researchers, the tool is aimed at countering the growing misuse of synthetic voices in digital spaces. It can help investigators confirm whether a voice recording belongs to a real speaker or has been artificially generated.
“This software addresses an urgent need in modern forensic science, where the rapid spread of AI-generated content challenges crime detection and evidence analysis,” Krishan said. Recently named among the world’s top two percent scientists by Stanford University, he is also the highest-ranked professor at Panjab University in 2025.
The research took six months and involved repeated trials to refine the model. Investigators expect the tool to ease the workload of forensic scientists, particularly in cases involving threatening calls, online fraud, cybercrime and misinformation campaigns.
While the current dataset remains small, Krishan noted that further work is planned. “The present version was trained on 100 samples. Pratibha intends to pursue her PhD to expand the dataset and enhance accuracy,” he said. The university will retain intellectual property rights over the software and is exploring potential industry collaborations for its broader application.
