small fixes and test fixes

TTS/bin/train_vocoder_wavernn.py

@@ -514,7 +514,7 @@ if __name__ == "__main__":
             new_fields["restore_path"] = args.restore_path
         new_fields["github_branch"] = get_git_branch()
         copy_model_files(
-            args.config_path, os.path.join(OUT_PATH, "config.json"), new_fields
+            c, args.config_path, OUT_PATH, new_fields
         )
         os.chmod(AUDIO_PATH, 0o775)
         os.chmod(OUT_PATH, 0o775)

TTS/tts/layers/speedy_speech/encoder.py

@@ -122,7 +122,7 @@ class Encoder(nn.Module):
                                      num_layers=3,
                                      dropout_p=0.5)
             # text encoder
-            self.encoder = Transformer(self.hidden_channels, **encoder_params)  # pylint: disable=unexpected-keyword-arg
+            self.encoder = RelativePositionTransformer(self.hidden_channels, **encoder_params)  # pylint: disable=unexpected-keyword-arg
         elif encoder_type.lower() == 'residual_conv_bn':
             self.pre = nn.Sequential(
                 nn.Conv1d(self.in_channels, self.hidden_channels, 1),

TTS/tts/models/speedy_speech.py

@@ -169,7 +169,6 @@ class SpeedySpeech(nn.Module):
             dr: [B, T_max]
             g: [B, C]
         """
-        breakpoint()
         o_en, o_en_dp, x_mask, g = self._forward_encoder(x, x_lengths, g)
         o_dr_log = self.duration_predictor(o_en_dp.detach(), x_mask)
         o_de, attn= self._forward_decoder(o_en, o_en_dp, dr, x_mask, y_lengths, g=g)

tests/inputs/test_glow_tts.json

@@ -86,7 +86,19 @@
     "warmup_steps": 4000,          // Noam decay steps to increase the learning rate from 0 to "lr"
     "seq_len_norm": false,         // Normalize eash sample loss with its length to alleviate imbalanced datasets. Use it if your dataset is small or has skewed distribution of sequence lengths.
 
-    "encoder_type": "transformer",
+    "hidden_channels_encoder": 192,
+        "hidden_channels_decoder": 192,
+        "hidden_channels_duration_predictor": 256,
+        "use_encoder_prenet": true,
+        "encoder_type": "rel_pos_transformer",
+        "encoder_params": {
+            "kernel_size":3,
+            "dropout_p": 0.1,
+            "num_layers": 6,
+            "num_heads": 2,
+            "hidden_channels_ffn": 768,
+            "input_length": null
+    },
 
     // TENSORBOARD and LOGGING
     "print_step": 25,       // Number of steps to log training on console.
@@ -107,6 +119,7 @@
     "max_seq_len": 500,     // DATASET-RELATED: maximum text length
     "compute_f0": false,     // compute f0 values in data-loader
     "compute_input_seq_cache": true,
+    "use_noise_augment": true,
 
     // PATHS
     "output_path": "tests/train_outputs/",

tests/inputs/test_speedy_speech.json

@@ -64,6 +64,8 @@
 
     // MODEL PARAMETERS
     "positional_encoding": true,
+    "hidden_channels": 128,
+    "encoder_type": "residual_conv_bn",
     "encoder_type": "residual_conv_bn",
     "encoder_params":{
         "kernel_size": 4,
@@ -71,13 +73,15 @@
         "num_conv_blocks": 2,
         "num_res_blocks": 13
     },
-    "decoder_residual_conv_bn_params":{
+    "decoder_type": "residual_conv_bn",
+    "decoder_params":{
         "kernel_size": 4,
         "dilations": [1, 2, 4, 8, 1, 2, 4, 8, 1, 2, 4, 8, 1, 2, 4, 8, 1],
         "num_conv_blocks": 2,
         "num_res_blocks": 17
     },
 
+
     // TRAINING
     "batch_size":64,       // Batch size for training. Lower values than 32 might cause hard to learn attention. It is overwritten by 'gradual_training'.
     "eval_batch_size":32,

tests/test_glow_tts.py

@@ -42,14 +42,22 @@ class GlowTTSTrainTest(unittest.TestCase):
         criterion = criterion = GlowTTSLoss()
 
         # model to train
-        model = GlowTts(num_chars=32,
+        model = GlowTts(
-                        hidden_channels=128,
+            num_chars=32,
-                        hidden_channels_ffn=32,
+            hidden_channels_enc=128,
+            hidden_channels_dec=128,
             hidden_channels_dp=32,
             out_channels=80,
-                        num_heads=2,
+            encoder_type='rel_pos_transformer',
-                        num_layers_enc=6,
+            encoder_params={
-                        dropout_p=0.1,
+                'kernel_size': 3,
+                'dropout_p': 0.1,
+                'num_layers': 6,
+                'num_heads': 2,
+                'hidden_channels_ffn': 768,  # 4 times the hidden_channels
+                'input_length': None
+            },
+            use_encoder_prenet=True,
             num_flow_blocks_dec=12,
             kernel_size_dec=5,
             dilation_rate=5,
@@ -58,25 +66,27 @@ class GlowTTSTrainTest(unittest.TestCase):
             num_speakers=0,
             c_in_channels=0,
             num_splits=4,
-                        num_sqz=1,
+            num_squeeze=1,
             sigmoid_scale=False,
-                        rel_attn_window_size=None,
+            mean_only=False).to(device)
-                        input_length=None,
-                        mean_only=False,
-                        hidden_channels_enc=None,
-                        hidden_channels_dec=None,
-                        use_encoder_prenet=False,
-                        encoder_type="transformer").to(device)
 
         # reference model to compare model weights
-        model_ref = GlowTts(num_chars=32,
+        model_ref = GlowTts(
-                            hidden_channels=128,
+            num_chars=32,
-                            hidden_channels_ffn=32,
+            hidden_channels_enc=128,
+            hidden_channels_dec=128,
             hidden_channels_dp=32,
             out_channels=80,
-                            num_heads=2,
+            encoder_type='rel_pos_transformer',
-                            num_layers_enc=6,
+            encoder_params={
-                            dropout_p=0.1,
+                'kernel_size': 3,
+                'dropout_p': 0.1,
+                'num_layers': 6,
+                'num_heads': 2,
+                'hidden_channels_ffn': 768,  # 4 times the hidden_channels
+                'input_length': None
+            },
+            use_encoder_prenet=True,
             num_flow_blocks_dec=12,
             kernel_size_dec=5,
             dilation_rate=5,
@@ -85,15 +95,9 @@ class GlowTTSTrainTest(unittest.TestCase):
             num_speakers=0,
             c_in_channels=0,
             num_splits=4,
-                            num_sqz=1,
+            num_squeeze=1,
             sigmoid_scale=False,
-                            rel_attn_window_size=None,
+            mean_only=False).to(device)
-                            input_length=None,
-                            mean_only=False,
-                            hidden_channels_enc=None,
-                            hidden_channels_dec=None,
-                            use_encoder_prenet=False,
-                            encoder_type="transformer").to(device)
 
         model.train()
         print(" > Num parameters for GlowTTS model:%s" %