more logging

voice_conversion/so-vits-svc-paul-v2/src/fakeyou_infer.py

@@ -196,6 +196,10 @@ def infer(
 ):
     """Inference"""
     #from .inference_main import infer
+
+    LOG.warning(f"Using device: {device}")
+    LOG.warning(f"Model path: {model_path}")
+
     from so_vits_svc_fork.inference.main import infer
 
     if not auto_predict_f0:

voice_conversion/so-vits-svc-paul-v2/src/so_vits_svc_fork/__main__.py

@@ -796,7 +796,7 @@ def clean():
     "-d",
     "--device",
     type=str,
-    default="cpu",
+    default="cuda",
     help="device to use",
 )
 def onnx(

voice_conversion/so-vits-svc-paul-v2/src/so_vits_svc_fork/inference/core.py

@@ -141,7 +141,7 @@ class Svc:
         _ = utils.load_checkpoint(self.net_g_path, self.net_g, None)
 
         LOG.info(f"Svc.load_model() utils.load_checkpoit() END")
-        
+
         _ = self.net_g.eval().to(self.device, dtype=self.dtype)
 
     def get_unit_f0(
@@ -154,12 +154,18 @@ class Svc:
             "crepe", "crepe-tiny", "parselmouth", "dio", "harvest"
         ] = "dio",
     ):
+        LOG.info(f"Svc.get_unit_f0() compute_f0 BEGIN")
+
         f0 = so_vits_svc_f0.compute_f0(
             audio,
             sampling_rate=self.target_sample,
             hop_length=self.hop_size,
             method=f0_method,
         )
+        LOG.info(f"Svc.get_unit_f0() compute_f0 END")
+
+        LOG.info(f"Svc.get_unit_f0() interpolate_f0 BEGIN")
+
         f0, uv = so_vits_svc_f0.interpolate_f0(f0)
         f0 = torch.as_tensor(f0, dtype=self.dtype, device=self.device)
         uv = torch.as_tensor(uv, dtype=self.dtype, device=self.device)
@@ -167,6 +173,10 @@ class Svc:
         f0 = f0.unsqueeze(0)
         uv = uv.unsqueeze(0)
 
+        LOG.info(f"Svc.get_unit_f0() interpolate_f0 END")
+
+        LOG.info(f"Svc.get_unit_f0() utils.get_content BEGIN")
+
         c = utils.get_content(
             self.hubert_model,
             audio,
@@ -174,15 +184,27 @@ class Svc:
             self.target_sample,
             self.contentvec_final_proj,
         ).to(self.dtype)
+
+        LOG.info(f"Svc.get_unit_f0() utils.get_content END")
+
+        LOG.info(f"Svc.get_unit_f0() utils.repeat_expand_2d BEGIN")
+
         c = utils.repeat_expand_2d(c.squeeze(0), f0.shape[1])
 
+        LOG.info(f"Svc.get_unit_f0() utils.repeat_expand_2d END")
+
         if cluster_infer_ratio != 0:
+
+            LOG.info(f"Svc.get_unit_f0() cluster.get_cluster_center_result BEGIN [ON CPU!!]")
+
             cluster_c = cluster.get_cluster_center_result(
                 self.cluster_model, c.cpu().numpy().T, speaker
             ).T
             cluster_c = torch.FloatTensor(cluster_c).to(self.device)
             c = cluster_infer_ratio * cluster_c + (1 - cluster_infer_ratio) * c
 
+            LOG.info(f"Svc.get_unit_f0() cluster.get_cluster_center_result END [ON CPU!!]")
+
         c = c.unsqueeze(0)
         return c, f0, uv
 

voice_conversion/so-vits-svc-paul-v2/src/so_vits_svc_fork/utils.py

@@ -29,19 +29,24 @@ HUBERT_SAMPLING_RATE = 16000
 
 
 def get_optimal_device(index: int = 0) -> torch.device:
+    LOG.warn("util.get_optimal_device()")
     if torch.cuda.is_available():
+        LOG.warn("util.get_optimal_device() ... cuda")
         return torch.device(f"cuda:{index % torch.cuda.device_count()}")
     elif torch.backends.mps.is_available():
+        LOG.warn("util.get_optimal_device() ... mps")
         return torch.device("mps")
     else:
         try:
             import torch_xla.core.xla_model as xm  # noqa
 
             if xm.xrt_world_size() > 0:
+                LOG.warn("util.get_optimal_device() ... xla")
                 return torch.device("xla")
             # return xm.xla_device()
         except ImportError:
             pass
+    LOG.warn("util.get_optimal_device() ... cpu")
     return torch.device("cpu")