Deep Learning based Crime Detection and Resource Creation Approach From Bengali Voice Calls

Tracking #: 735-1715


Nahid HossainORCID logo
Khalid Saifullah
Mohammad Masudul Alam
Prosenjit Majumder Joy
Jadir Ibna Hasan
Salekul Islam

Responsible editor: 

Bharathi Raja Chakravarthi

Submission Type: 

Research Paper


Mobile phones have revolutionized our way of communication. Despite its numerous benefits, it has become a great utility for conducting crimes or making threats. Due to the large number of users it is almost impossible for security forces to take proactive measures against those crimes. In this paper, with the help of machine learning, we focus on building a system that can detect potential threats in phone calls. We develop (to the best of our knowledge) the very first Bengali voice call dataset to train the machine learning system. Our system takes a voice call and uses a Deep 1D Convolutional Neural Network to analyze the call and a Multi-Layer Perceptron to decide whether any threats exist or not. The proposed simple baseline solution, trained on our $\sim$9hrs. worth voice call dataset, is able to achieve $91$\% precision, recall and F1-score in detecting the crime calls. We believe, in future these systems will aid in assisting in investigations, evaluating voice conversations, and giving predictions and estimations for potential threats. All of our recorded calls are freely available to use by the future researchers at:



  • Under Review

Data repository URLs: 

Date of Submission: 

Wednesday, November 30, 2022