ArcticEcho: A Speaker-Controlled Voice Cloning Dataset for Modern Deepfake Detection Benchmarking

ArcticEcho is a controlled voice cloning dataset containing 24,752 audio files (21.04 hours) across 18 speakers from the CMU Arctic corpus. The dataset maintains perfect speaker-content correspondence between real and synthetic audio, generated using state-of-the-art voice cloning systems (ElevenLabs, Metavoice, OpenVoice). Dataset Structure:- real.zip: Original CMU Arctic audio files organized by speaker- fake.zip: Synthetic audio generated with identical speaker-content mapping File Formats: 84.2% PCM WAV (16kHz), 15.8% MP3Speakers: 18 (diverse gender, age, accent characteristics)Applications: Audio deepfake detection, voice cloning research, security applications --- Version 02 has all audio standardized to 16kHz, Mono WAV.