JoeyHeisenberg

Train failed because the loss is nan

27

I train the network on pytorch1.5 with 3 gpu, and train failed because the loss is nan hparams ################################ # Optimization Hyperparameters # ################################ use_saved_learning_rate=False, learning_rate=1e-3, weight_decay=1e-6, grad_clip_thresh=1.0, batch_size=24, mask_padding=True,...

A bug in HiFiGANScaleDiscriminator

2

seems that weight norm doesn't work in HiFiGANScaleDiscriminator, in which we only get conv1d https://github.com/espnet/espnet/blob/f274ebed88e3b4820b23ea71fb7f9f6d56706be9/espnet2/gan_tts/hifigan/hifigan.py#L410-L573

关于Embedding的问题

3

你好，看了里面的实现，你是将汉字_pos的特征作为整体进行embedding，和论文中将汉字、词性、是否多音字三种特征分别embedding后再concate的方法不同，有比较过之间的差异么？