Machine Learning-Based Speech Recognition of Chinese Dialects Method for Mobile Forensics

Abstract

The software of QQ, WeChat and other voice chat tools become increasingly popular, people can use these methods to communicate expediently. In the meantime, criminals also employ these speech communication tools to engage in criminal activities. These voice messages will include in the smartphones, and the speech data are important for the law enforcement officers. China has large population, criminal gangs always have local characteristics and dialects are a common way of communication between criminals. So, lots of speech date will store in the mobile phones. In the work of mobile forensics, investigators should take much time and human effort to recognition and analysis the dialects voice data. For this reason, the law enforcement personnel need to use intelligent method to investigation local dialects voice data in the mobile phone forensics efficiently. In this paper, we propose a mobile forensics method that based on LSTM (Long Short-Term Memory) neural network technique to automatically recognize dialects voice data for investigation and forensics.