TalkNet: Fully-Convolutional Non-Autoregressive Speech Synthesis Model

arxiv.org/abs/2005.05514

Stanislav Beliaev, Yurii Rebryk, Boris Ginsburg

LJSpeech

Unseen samples (test split), choosing random 10 examples.

Ground Truth GT + WaveGlow Tacotron2 + WaveGlow TalkNet + WaveGlow
LJ050-0118
LJ048-0033
LJ016-0192
LJ002-0253
LJ010-0257
LJ008-0121
LJ009-0106
LJ013-0104
LJ020-0092
LJ016-0314