This forked project integrates the Progressive Enhancement Learning (PEL) framework from the paper Exploiting Fine-grained Face Forgery Clues via Progressive Enhancement Learning into a synthetic speech detection system. By using LFCC and CQT as features, we enhance the detection of subtle forgery clues in synthetic speech. The implementation includes self-enhancement and mutual-enhancement modules to progressively refine feature learning, inspired by the techniques used in the PEL framework for face forgery detection.

Implementation of Progressive Enhancement Learning for Audio Deepfake Detection

This repository contains an implementation of Progressive Enhancement Learning applied to the detection of audio deepfakes.

Directory Structure

Progressive_Enhancement_Learning/: This directory houses the primary scripts for this project.
- demo1.py: The original implementation of the makaijie/End-to-End-Dual-Branch-Network-Towards-Synthetic-Speech-Detection repository.
- demo2.py: An implementation incorporating Self and Mutual Enhancement based on the paper.

Current Status

The code has been tested on single audio samples.
The full training pipeline will be updated soon.

End-to-End Dual-Branch Network Towards Synthetic Speech Detection

Prerequisites

NVIDIA GPU+CUDA CuDNN
Install Torch1.8 and dependencies

Training and Test Details

Please adjust the file location before training and testing;
Data Preparation
- Change the Feature Engineering/CQT/cqt_extract.py, Feature Engineering/LFCC/extract_lfcc.m and Feature Engineering/LFCC/reload_data.py
- Run the Feature Engineering/CQT/cqt_extract.py, Feature Engineering/LFCC/extract_lfcc.m and Feature Engineering/LFCC/reload_data.py
When you train the network
- Change the dual-branch_sum_loss.py or dual-branch_alternative_loss.py
- Run the dual-branch_sum_loss.py or dual-branch_alternative_loss.py
When you test the network
- Change the Result_sum_loss/test_dual.py or Result_alternative_loss/test_dual.py
- Run the Result_sum_loss/test_dual.py or Result_alternative_loss/test_dual.py

Acknowledgements

The code of this work is adapted from https://github.com/yzyouzhang/AIR-ASVspoof, https://github.com/yzyouzhang/Empirical-Channel-CM and https://github.com/joaomonteirof/e2e_antispoofing.