Merge pull request #858 from PaddlePaddle/ctc

ctc using nn.Dropout; ds2 libri vald batch_size / 4

deepspeech/exps/deepspeech2/model.py

@@ -235,7 +235,7 @@ class DeepSpeech2Trainer(Trainer):
             num_workers=config.collator.num_workers)
         self.valid_loader = DataLoader(
             dev_dataset,
-            batch_size=config.collator.batch_size,
+            batch_size=int(config.collator.batch_size / 4),
             shuffle=False,
             drop_last=False,
             collate_fn=collate_fn_dev)

deepspeech/modules/ctc.py

@@ -56,7 +56,7 @@ class CTCDecoder(nn.Layer):
 
         self.blank_id = blank_id
         self.odim = odim
-        self.dropout_rate = dropout_rate
+        self.dropout = nn.Dropout(dropout_rate)
         self.ctc_lo = nn.Linear(enc_n_units, self.odim)
         reduction_type = "sum" if reduction else "none"
         self.criterion = CTCLoss(
@@ -79,7 +79,7 @@ class CTCDecoder(nn.Layer):
         Returns:
             loss (Tenosr): ctc loss value, scalar.
         """
-        logits = self.ctc_lo(F.dropout(hs_pad, p=self.dropout_rate))
+        logits = self.ctc_lo(self.dropout(hs_pad))
         loss = self.criterion(logits, ys_pad, hlens, ys_lens)
         return loss
 

deepspeech/modules/loss.py

@@ -45,9 +45,9 @@ class CTCLoss(nn.Layer):
         logger.info(f"CTCLoss Grad Norm Type: {grad_norm_type}")
         if grad_norm_type == 'instance':
             self.norm_by_times = True
-        if grad_norm_type == 'batch':
+        elif grad_norm_type == 'batch':
             self.norm_by_batchsize = True
-        if grad_norm_type == 'frame':
+        elif grad_norm_type == 'frame':
             self.norm_by_total_logits_len = True
 
     def forward(self, logits, ys_pad, hlens, ys_lens):

examples/librispeech/s1/conf/transformer.yaml

@@ -8,7 +8,7 @@ data:
   min_output_len: 0.0 # tokens
   max_output_len: 400.0 # tokens
   min_output_input_ratio: 0.05
-  max_output_input_ratio: 10.0
+  max_output_input_ratio: 100.0
 
 collator:
   vocab_filepath: data/vocab.txt
@@ -16,7 +16,7 @@ collator:
   spm_model_prefix: 'data/bpe_unigram_5000'
   mean_std_filepath: ""
   augmentation_config: conf/augmentation.json
-  batch_size: 64
+  batch_size: 32
   raw_wav: True  # use raw_wav or kaldi feature
   specgram_type: fbank #linear, mfcc, fbank
   feat_dim: 80
@@ -75,7 +75,7 @@ model:
 
 training:
   n_epoch: 120 
-  accum_grad: 2
+  accum_grad: 4
   global_grad_clip: 5.0
   optim: adam
   optim_conf: