VoiceWukong+EnvSDD Mixed Deepfake Audio Dataset with Train/Valid/Test Splits

Mixed audio dataset for deepfake voice detection in noisy environments. Deepfake voice (VoiceWukong expfade20/en) mixed with real/fake environmental backgrounds (EnvSDD).Split design (no data leakage):- VoiceWukong fake_voice: random 7:2:1 (seed=42), 2000 files selected (train=1400 / valid=400 / test=200)- EnvSDD real_bg: official split (train=683 / valid=95 / test=222)- EnvSDD fake_bg: official split (train=717 / valid=105 / test=178)6 categories = 2 bg types x 3 volume levels:fake_voice_real_bg, fake_voice_fake_bgVolume: balanced (60/60 dB), voice_dominant (80/60 dB), bg_dominant (60/80 dB) [REF_DB=100]Contents:- wukong_envsdd_code.zip: synthesis script + EnvSDD backgrounds.csv + all split manifests (wukong_fake_voice.csv, envsdd_real_bg.csv, envsdd_fake_bg.csv) + dataset_splits_summary.csv- wukong_envsdd_2s.zip: 12066 WAV files, 2-second, 16kHz mono PCM16- wukong_envsdd_4s_fake_voice_real_bg.zip: 6033 WAV files, 4-second, 16kHz mono PCM16- wukong_envsdd_4s_fake_voice_fake_bg.zip: 6033 WAV files, 4-second, 16kHz mono PCM16