ericwudayi

I try notebook E7 exercise, and try to resynthsize the singing voice, and I found that no matter how I tune the parameter, it seems cannot have a good resynthesis...

Good recover from real melspectrogram to audio, but problem in autoencoder.

3

Melgan vocoder can convert the real melspectrogram to waveform well, but I find that there is a problem when melgan vocoder convert the "model output melspectrogram". The melspectrogram extracted from...

Local execution with compiled graph(#4080)

3

# TL;DR Support user-defined execution sequence in local pyfylte. Now subworkflow/tasks can be sequence by `>>` operation within workflow. for example: ``` @workflow() def wf(): a = print_a() # task:...

Add update_execution func in remote

5

# TL;DR As Title ## Type - [ ] Bug Fix - [x] Feature - [ ] Plugin ## Are all requirements met? - [x] Code completed - [ ]...

# TL;DR _Please replace this text with a description of what this PR accomplishes._ ## Type - [ ] Bug Fix - [x] Feature - [ ] Plugin ## Are...

Bert 加入以后, 底模训练, KL loss爆炸.

2

感谢大大提供这么好的repo, 我这边在用训练一个基底模型, 用版主預先訓練好的prosody.pt 來做character embedding的提取, 然後把提取出來的feature跟chinses_ipa出來的音素做對齊. 這樣算出來的feature, 每一個時間點的norm大約在10左右 不知道在加入bert feature以后有没有发现KL爆炸的问题? 我對Bert feature加入很多normalization都沒有辦法解決這個問題. 不知道版主有沒有遇到過? ``` def forward(self, x, x_lengths, bert, bert_legnths): bert_emb = self.tanh(self.bert_proj(self.tanh(bert)).transpose(1, 2)) #bert_emb = bert_emb / (torch.norm(bert_emb, dim=-1).unsqueeze(-1)...