How to correctly use CTC Loss with GRU in pytorch?

Question

I am trying to create ASR and I am still learning so, I am just trying with a simple GRU:

MySpeechRecognition(
  (gru): GRU(128, 128, num_layers=5, batch_first=True, dropout=0.5)
  (dropout): Dropout(p=0.3, inplace=False)
  (fc1): Linear(in_features=128, out_features=512, bias=True)
  (fc2): Linear(in_features=512, out_features=28, bias=True)
)

Classifies each output as one of the possible alphabets + space + blank.

Then I use CTC Loss Function and Adam optimizer:

lr = 5e-4
criterion = nn.CTCLoss(blank=28, zero_infinity=False)
optimizer = torch.optim.Adam(net.parameters(), lr=lr)

In my training loop (I am only showing the problematic area):

output, h = mynet(specs, h)
print(output.size())
output = F.log_softmax(output, dim=2)
output = output.transpose(0,1)
# calculate the loss and perform backprop
loss = criterion(output, labels, input_lengths, label_lengths)
loss.backward()

I get this error:

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
<ipython-input-133-5e47e7b03a46> in <module>
     42         output = output.transpose(0,1)
     43         # calculate the loss and perform backprop
---> 44         loss = criterion(output, labels, input_lengths, label_lengths)
     45         loss.backward()
     46         # `clip_grad_norm` helps prevent the exploding gradient problem in RNNs / LSTMs.

/opt/conda/lib/python3.7/site-packages/torch/nn/modules/module.py in __call__(self, *input, **kwargs)
    548             result = self._slow_forward(*input, **kwargs)
    549         else:
--> 550             result = self.forward(*input, **kwargs)
    551         for hook in self._forward_hooks.values():
    552             hook_result = hook(self, input, result)

/opt/conda/lib/python3.7/site-packages/torch/nn/modules/loss.py in forward(self, log_probs, targets, input_lengths, target_lengths)
   1309     def forward(self, log_probs, targets, input_lengths, target_lengths):
   1310         return F.ctc_loss(log_probs, targets, input_lengths, target_lengths, self.blank, self.reduction,
-> 1311                           self.zero_infinity)
   1312 
   1313 # TODO: L1HingeEmbeddingCriterion

/opt/conda/lib/python3.7/site-packages/torch/nn/functional.py in ctc_loss(log_probs, targets, input_lengths, target_lengths, blank, reduction, zero_infinity)
   2050     """
   2051     return torch.ctc_loss(log_probs, targets, input_lengths, target_lengths, blank, _Reduction.get_enum(reduction),
-> 2052                           zero_infinity)
   2053 
   2054 

RuntimeError: blank must be in label range

I am not sure why I am getting this error. I tried changing to

labels.float()

Thanks.

score 4 · Accepted Answer · answered Jun 08 '20 at 06:17

Your model predicts 28 classes, therefore the output of the model has size [batch_size, seq_len, 28] (or [seq_len, batch_size, 28] for the log probabilities that are given to the CTC loss). In the nn.CTCLoss you set blank=28, which means that the blank label is the class with index 28. To get the log probabilities for the blank label you would index it as output[:, :, 28], but that doesn't work, because that index is out of range, as the valid indices are 0 to 27.

The last class in your output is at index 27, hence it should be blank=27:

criterion = nn.CTCLoss(blank=27, zero_infinity=False)

Thank you so much!! I was so frustrated with this. It worked and now I understand it better :) — swe87, Jun 08 '20 at 08:31

How to correctly use CTC Loss with GRU in pytorch?

1 Answers1