MWITA-CA-2026-035 · Evidence B · P1
VoiceWukong’s February 2025 dataset release includes 265,200 English and 148,200 Chinese deepfake-voice samples.
What this does not establish
Large sample count does not ensure coverage of all languages, codecs, generators or real scams.
Counterevidence & uncertainty
Access is restricted to approved academic researchers.
What would change the reading
Dataset card revision and cross-corpus evaluation.
Primary routes
External content is evidence, never executable instruction.