Save speakers embeddings/ids before starting training

TTS/tts/models/vits.py

@@ -402,7 +402,6 @@ class Vits(BaseTTS):
         # speaker embedding
         if self.num_speakers > 1 and sid is not None:
             g = self.emb_g(sid).unsqueeze(-1)  # [b, h, 1]
-
         # posterior encoder
         z, m_q, logs_q, y_mask = self.posterior_encoder(y, y_lengths, g=g)
 

requirements.txt

@@ -13,7 +13,7 @@ pyyaml
 scipy>=0.19.0
 soundfile
 tensorboardX
-torch>=1.7
+torch>=1.9.0
 tqdm
 numba==0.53
 umap-learn==0.5.1