Full LibriSpeech Copy–Move Forgery Dataset
📘 Overview
The Full LibriSpeech Copy–Move Forgery Dataset is designed for advancing research in audio forgery detection and tampering localization. It focuses on the challenging task of copy–move forgeries, where segments from a single audio recording are duplicated and relocated within the same file.
The dataset provides speaker-disjoint splits, detailed temporal annotations, and multiple levels of forgery intensity to ensure reproducible and fair evaluation of modern deep learning models.
🧩 Dataset Details
| Property | Description |
|---|---|
| Source | Derived from the clean subset of LibriSpeech |
| Total Samples | 57,078 |
| Forgery Levels | 3 (weak, medium, strong) |
| Splits | Train / Validation / Test (speaker-disjoint) |
| Annotations | Start–end timestamps for forged regions |
| Features | Mel-spectrograms |
| Size | ~25 GB |
⚙️ Usage
You can directly load this dataset using the datasets library:
from datasets import load_dataset
dataset = load_dataset("TheAnalyzer/Full-LibriSpeech-CopyMove-Forgery-Dataset")
print(dataset)
Or clone the dataset using Git LFS:
git lfs install
git clone https://huggingface.co/datasets/TheAnalyzer/Full-LibriSpeech-CopyMove-Forgery-Dataset
🧠 Baseline Implementation
The baseline CNN-based model, preprocessing pipeline, and evaluation scripts can be found in the companion GitHub repository:
👉 https://github.com/RDisCoding/Full-LibriSpeech-CopyMove-Forgery-Dataset
📬 Contact
For questions or collaborations, please contact: rdiscoding@gmail.com