Detecting Forged Audio Files Using "Mixed Paste" Command: A Deep Learning Approach Based on Korean Phonemic Features

Area of Science:

• Digital Forensics
• Artificial Intelligence
• Audio Engineering

Background:

• Voice recordings are increasingly used as digital evidence in legal cases.
• Allegations of audio file forgery are rising, necessitating robust authentication methods.
• Existing forensic techniques struggle to detect sophisticated audio manipulations like "Mixed Paste".

Purpose of the Study:

• To develop and validate a deep learning methodology for identifying forged voice recordings.
• To specifically address the detection of audio files altered using the "Mixed Paste" technique.
• To enhance the reliability of digital audio evidence in legal contexts.

Main Methods:

• A hybrid deep learning model combining Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) was developed.
• Features were extracted from spectrograms and sequences of Korean consonant types for analysis.
• The model was trained on a dataset of authentic and "Mixed Paste" forged audio recordings from iPhones.

Main Results:

• The hybrid deep learning model achieved a high accuracy rate of 97.5% in detecting forged audio files.
• The model's effectiveness was consistent across different smartphone models and audio editing software.
• Validation tests confirmed the model's capability to identify manipulated audio files.

Conclusions:

• The proposed deep learning framework offers a powerful new tool for audio file authentication in digital forensics.
• This research advances audio forensics by introducing a novel hybrid model capable of detecting "Mixed Paste" audio forgeries.
• The findings support the broader application of AI in ensuring the integrity of digital evidence.