Sound examples

Comparison of different systems, in all cases the input mel-spectrogram/WORLD features are obtained by analysis of a recording.

GT = Reference recording

NW = Neural WORLD (our proposed system)

NW-NoAdv = Neural WORLD (our proposed system) without adversarial training

E-PWG = Excited Parallel WaveGAN (non-autoregressive baseline)

WORLD = WORLD vocoder (signal processing based)

AR-WNV = Autoregressive WaveNet vocoder

Seen singers

Unseen singers