Medix-AI/coqui-tts
Medix-AI
/
coqui-tts
mirror of https://github.com/coqui-ai/TTS.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Fix bug in Graves Attn 

On my machine at Graves attention the variable self.J ( self.J = torch.arange(0, inputs.shape[1]+2).to(inputs.device) + 0.5) is a LongTensor, but it must be a float tensor. So I get the following error:

Traceback (most recent call last):
  File "train.py", line 704, in <module>
    main(args)
  File "train.py", line 619, in main
    global_step, epoch)
  File "train.py", line 170, in train
    text_input, text_lengths, mel_input, speaker_embeddings=speaker_embeddings)
  File "/home/edresson/anaconda3/envs/TTS2/lib/python3.6/site-packages/torch/nn/modules/module.py", line 489, in __call__
    result = self.forward(*input, **kwargs)
  File "/mnt/edresson/DD/TTS/voice-clonning/TTS/tts_namespace/TTS/models/tacotron.py", line 121, in forward
    self.speaker_embeddings_projected)
  File "/home/edresson/anaconda3/envs/TTS2/lib/python3.6/site-packages/torch/nn/modules/module.py", line 489, in __call__
    result = self.forward(*input, **kwargs)
  File "/mnt/edresson/DD/TTS/voice-clonning/TTS/tts_namespace/TTS/layers/tacotron.py", line 435, in forward
    output, stop_token, attention = self.decode(inputs, mask)
  File "/mnt/edresson/DD/TTS/voice-clonning/TTS/tts_namespace/TTS/layers/tacotron.py", line 367, in decode
    self.attention_rnn_hidden, inputs, self.processed_inputs, mask)
  File "/home/edresson/anaconda3/envs/TTS2/lib/python3.6/site-packages/torch/nn/modules/module.py", line 489, in __call__
    result = self.forward(*input, **kwargs)
  File "/mnt/edresson/DD/TTS/voice-clonning/TTS/tts_namespace/TTS/layers/common_layers.py", line 180, in forward
    phi_t = g_t.unsqueeze(-1) * (1.0 / (1.0 + torch.sigmoid((mu_t.unsqueeze(-1) - j) / sig_t.unsqueeze(-1))))
RuntimeError: expected type torch.cuda.FloatTensor but got torch.cuda.LongTensor


In addition the + 0.5 operation is canceled if it is a LongTensor.
Test: 
>>> torch.arange(0, 10) 
tensor([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
>>> torch.arange(0, 10) + 0.5
tensor([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
>>> torch.arange(0, 10.0) + 0.5
tensor([0.5000, 1.5000, 2.5000, 3.5000, 4.5000, 5.5000, 6.5000, 7.5000, 8.5000,
        9.5000])

To resolve this I forced the arrange range to float:
self.J = torch.arange(0, inputs.shape[1]+2.0).to(inputs.device) + 0.5
This commit is contained in:
Edresson Casanova 2020-05-04 17:52:58 -03:00 committed by GitHub
parent f7b1cad9ee
commit cce13ee245
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
layers/common_layers.py
View File

@ -138,7 +138,7 @@ class GravesAttention(nn.Module):
def init_states(self, inputs):
if self.J is None or inputs.shape[1]+1 > self.J.shape[-1]:
self.J = torch.arange(0, inputs.shape[1]+2).to(inputs.device) + 0.5
self.J = torch.arange(0, inputs.shape[1]+2.0).to(inputs.device) + 0.5
self.attention_weights = torch.zeros(inputs.shape[0], inputs.shape[1]).to(inputs.device)
self.mu_prev = torch.zeros(inputs.shape[0], self.K).to(inputs.device)