This project implements and trains a Countermeasure (CM) system to detect spoofing attacks (replays, synthetic speech, etc.) on the Logical Access (LA) partition of the ASVSpoof 2019 Dataset. The core architecture is a LightCNN (LCNN), implemented from scratch in PyTorch following the specifications from papers:
- STC Antispoofing Systems for the ASVspoof2019 Challenge by Speech Technology Center - used as the foundation for the LightCNN architecture implementation
- A Comparative Study on Recent Neural Spoofing Countermeasures for Synthetic Speech Detection - used for the training recipe and data preparation scheme
These instructions will give you a copy of the project up and running on your local machine for development and training purposes.
A step-by-step series of examples to get a development environment running.
-
Create and activate a Conda environment:
conda create -n voice-anti-spoofing python=3.8 conda activate voice-anti-spoofing
-
Clone this repository and navigate into the project directory:
git clone https://github.com/ilyalychagin/Voice_Anti-spoofing.git cd Voice_Anti-spoofing -
Install the required Python packages:
pip install -r requirements.txt
-
Download the Dataset:
- Download the
LApartition from the official ASVSpoof 2019 dataset on Kaggle. - Place the downloaded data in the root of the project directory.
- Download the
-
Start the training process:
python3 train.py
The script will begin training the LightCNN model on the ASVSpoof 2019 LA training set.
EER: 7% (Equal error rate)
- Acknowledgement to the pytorch_project_template for providing a structured project foundation.