TTS & cuda processing updated

cuda.py

@@ -0,0 +1,5 @@
+import torch
+
+print("Torch version:",torch.__version__)
+
+print("Is CUDA enabled?",torch.cuda.is_available())

src/blackbox/tts.py

@@ -1,4 +1,5 @@
 import io
+import time
 from ntpath import join
 
 from fastapi import Request, Response, status
@@ -16,7 +17,9 @@ class TTS(Blackbox):
 
     def processing(self, *args, **kwargs) -> io.BytesIO:
         text = args[0]
+        current_time = time.time()
         audio = self.tts_service.read(text)
+        print("#### TTS Service consume : ", (time.time()-current_time))
         return audio
 
     def valid(self, *args, **kwargs) -> bool:

src/tts/tts_service.py

@@ -53,7 +53,7 @@ class TTService():
             len(symbols),
             self.hps.data.filter_length // 2 + 1,
             self.hps.train.segment_size // self.hps.data.hop_length,
-            **self.hps.model).cpu()
+            **self.hps.model).cuda()
         _ = self.net_g.eval()
         _ = utils.load_checkpoint(cfg["model"], self.net_g, None)
 
@@ -69,8 +69,8 @@ class TTService():
         stn_tst = self.get_text(text, self.hps)
         
         with torch.no_grad():
-            x_tst = stn_tst.cpu().unsqueeze(0)
-            x_tst_lengths = torch.LongTensor([stn_tst.size(0)]).cpu()
+            x_tst = stn_tst.cuda().unsqueeze(0)
+            x_tst_lengths = torch.LongTensor([stn_tst.size(0)]).cuda()
             # tp = self.net_g.infer(x_tst, x_tst_lengths, noise_scale=.667, noise_scale_w=0.2, length_scale=self.speed)
             audio = self.net_g.infer(x_tst, x_tst_lengths, noise_scale=.667, noise_scale_w=0.2, length_scale=self.speed)[0][
                 0, 0].data.cpu().float().numpy()

src/tts/vits/monotonic_align/__init__.py

@@ -1,6 +1,6 @@
 import numpy as np
 import torch
-from .monotonic_align.core import maximum_path_c
+from .core import maximum_path_c
 
 
 def maximum_path(neg_cent, mask):