Benchmark-ready packaging of the test split of LibriSeVoc — a vocoder
re-synthesis corpus for synthetic / deepfake voice detection.
LibriSeVoc (Sun et al., arXiv:2304.13085) takes genuine LibriTTS utterances and
re-synthesises each one with six neural vocoders — DiffWave, MelGAN, Parallel
WaveGAN, WaveGrad, WaveNet, and WaveRNN. The detection task is binary:
bonafide (the genuine LibriTTS audio) vs. spoof (a vocoded re-synthesis).… See the full description on the dataset page:
https://huggingface.co/datasets/SpeechAntiSpoofingBenchmarks/LibriSeVoc.
