Datasets:
issf
/

Modalities:
Audio
Text
Formats:
parquet
Languages:
English
ArXiv:
Libraries:
Datasets
Dask
License:

Acknowledge license to accept the repository

Our team may take 2-3 days to process your request. Please use an email address that is from your current affiliation, not personal email addresses (e.g., gmail). To avoid anonymous downloads, we accept officially affiliated email addresses.

By accepting and downloading the dataset, you agree to the license and restrictions. Please check https://datasets.issflab.net/Famous_fig.html for full details. We may send important updates or license-related information to the registered e-mail address.

Log in or Sign Up to review the conditions and access this dataset content.

Famous Figures: Collecting, Curating, and Annotating Good Quality Speech Deepfake Dataset for Famous Figures — Process and Challenges

Paper, Project Page

Interspeech 2025

Abstract

Current audio deepfake detection systems fail to protect specific individuals from targeted voice spoofing attacks.A comprehensive methodology for collecting, curating, and generating high-quality speech datasets specifically designed to protect political figures.

Salient Features of our Dataset

✓ Speaker-specific protection

✓ 8-second average duration

✓ Transcription-based segmentation

✓ Natural sentence boundaries

✓ Superior synthesis quality

image/png

Acknowledgement

This work uses only public‑domain or licensed speech sources of public figures. Audio clips are included for research/educational demonstration; rights remain with original owners. Please see the paper for data collection ethics and licensing details.

Downloads last month
22