num_speakers larger than 1

models/tacotron.py

@@ -29,7 +29,7 @@ class Tacotron(nn.Module):
         self.linear_dim = linear_dim
         self.embedding = nn.Embedding(num_chars, 256)
         self.embedding.weight.data.normal_(0, 0.3)
-        if num_speakers > 0:
+        if num_speakers > 1:
             self.speaker_embedding = nn.Embedding(num_speakers, 256)
             self.speaker_embedding.weight.data.normal_(0, 0.3)
         self.encoder = Encoder(256)

models/tacotron2.py

@@ -29,7 +29,7 @@ class Tacotron2(nn.Module):
         std = sqrt(2.0 / (num_chars + 512))
         val = sqrt(3.0) * std  # uniform bounds for std
         self.embedding.weight.data.uniform_(-val, val)
-        if num_speakers > 0:
+        if num_speakers > 1:
             self.speaker_embedding = nn.Embedding(num_speakers, 512)
             self.speaker_embedding.weight.data.normal_(0, 0.3)
         self.encoder = Encoder(512)

models/tacotrongst.py

@@ -30,7 +30,7 @@ class TacotronGST(nn.Module):
         self.linear_dim = linear_dim
         self.embedding = nn.Embedding(num_chars, 256)
         self.embedding.weight.data.normal_(0, 0.3)
-        if num_speakers > 0:
+        if num_speakers > 1:
             self.speaker_embedding = nn.Embedding(num_speakers, 256)
             self.speaker_embedding.weight.data.normal_(0, 0.3)
         self.encoder = Encoder(256)

train.py

@@ -78,7 +78,7 @@ def setup_loader(is_val=False, verbose=False):
 def train(model, criterion, criterion_st, optimizer, optimizer_st, scheduler,
           ap, epoch):
     data_loader = setup_loader(is_val=False, verbose=(epoch==0))
-    if c.num_speakers > 0:
+    if c.num_speakers > 1:
         speaker_mapping = load_speaker_mapping(OUT_PATH)
     model.train()
     epoch_time = 0
@@ -102,7 +102,7 @@ def train(model, criterion, criterion_st, optimizer, optimizer_st, scheduler,
         avg_text_length = torch.mean(text_lengths.float())
         avg_spec_length = torch.mean(mel_lengths.float())
 
-        if c.num_speakers > 0:
+        if c.num_speakers > 1:
             speaker_ids = []
             for speaker_name in speaker_names:
                 if speaker_name not in speaker_mapping:
@@ -272,14 +272,14 @@ def train(model, criterion, criterion_st, optimizer, optimizer_st, scheduler,
             tb_logger.tb_model_weights(model, current_step)
 
     # save speaker mapping
-    if c.num_speakers > 0:
+    if c.num_speakers > 1:
         save_speaker_mapping(OUT_PATH, speaker_mapping)
     return avg_postnet_loss, current_step
 
 
 def evaluate(model, criterion, criterion_st, ap, current_step, epoch):
     data_loader = setup_loader(is_val=True)
-    if c.num_speakers > 0:
+    if c.num_speakers > 1:
         speaker_mapping = load_speaker_mapping(OUT_PATH)
     model.eval()
     epoch_time = 0
@@ -311,7 +311,7 @@ def evaluate(model, criterion, criterion_st, ap, current_step, epoch):
                 mel_lengths = data[5]
                 stop_targets = data[6]
 
-                if c.num_speakers > 0:
+                if c.num_speakers > 1:
                     speaker_ids = [speaker_mapping[speaker_name]
                                    for speaker_name in speaker_names]
                     speaker_ids = torch.LongTensor(speaker_ids)
@@ -415,7 +415,7 @@ def evaluate(model, criterion, criterion_st, ap, current_step, epoch):
         test_audios = {}
         test_figures = {}
         print(" | > Synthesizing test sentences")
-        speaker_id = 0 if c.num_speakers > 0 else None
+        speaker_id = 0 if c.num_speakers > 1 else None
         for idx, test_sentence in enumerate(test_sentences):
             try:
                 wav, alignment, decoder_output, postnet_output, stop_tokens = synthesis(
@@ -484,7 +484,7 @@ def main(args):
         args.restore_step = checkpoint['step']
         # copying speakers.json
         prev_out_path = os.path.dirname(args.restore_path)
-        if c.num_speakers > 0:
+        if c.num_speakers > 1:
             copy_speaker_mapping(prev_out_path, OUT_PATH)
     else:
         args.restore_step = 0