Update app.py

app.py

@@ -36,6 +36,7 @@ theme = gr.themes.Base(
     font=[gr.themes.GoogleFont('Libre Franklin'), gr.themes.GoogleFont('Public Sans'), 'system-ui', 'sans-serif'],
 )
 
+from Modules.diffusion.sampler import DiffusionSampler, ADPM2Sampler, KarrasSchedule
 
 voicelist = ['VO_JA_Kamisato_Ayaka_About_Kujou_Sara','hontonokimochi','gaen_original']
 voices = {}
@@ -89,7 +90,22 @@ def synthesize(text, voice, lngsteps, password, progress=gr.Progress()):
 #         return (24000, np.concatenate(audios))
 #     else:
 #         raise gr.Error('Wrong access code')
+    
 def clsynthesize(text, voice, vcsteps, embscale, alpha, beta, progress=gr.Progress()):
+
+
+    torch.manual_seed(0)
+    torch.backends.cudnn.benchmark = False
+    torch.backends.cudnn.deterministic = True
+
+
+    random.seed(0)
+
+    sampler = DiffusionSampler(
+        model.diffusion.diffusion,
+        sampler=ADPM2Sampler(),
+        sigma_schedule=KarrasSchedule(sigma_min=0.0001, sigma_max=0.75, rho=4.), # empirical parameters
+        clamp=False)
     # if text.strip() == "":
     #     raise gr.Error("You must enter some text")
     # # if global_phonemizer.phonemize([text]) > 300:
@@ -120,12 +136,28 @@ def clsynthesize(text, voice, vcsteps, embscale, alpha, beta, progress=gr.Progre
 
     
 def ljsynthesize(text, steps,embscale, progress=gr.Progress()):
+
+
+
+    torch.manual_seed(0)
+    torch.backends.cudnn.benchmark = False
+    torch.backends.cudnn.deterministic = True
+
+
+    random.seed(0)
+
+    sampler = DiffusionSampler(
+        model.diffusion.diffusion,
+        sampler=ADPM2Sampler(),
+        sigma_schedule=KarrasSchedule(sigma_min=0.0001, sigma_max=0.75, rho=4.), # empirical parameters
+        clamp=False)
+    
     # if text.strip() == "":
     #     raise gr.Error("You must enter some text")
     # # if global_phonemizer.phonemize([text]) > 300:
     # if len(text) > 400:
     #     raise gr.Error("Text must be under 400 characters")
-    noise = torch.randn(1,1,256).to('cuda' if torch.cuda.is_available() else 'cpu')
+    noise = torch.tanh(torch.randn(1,1,256).to('cuda' if torch.cuda.is_available() else 'cpu'))
     # return (24000, Text-guided Inferenceimportable.inference(text, noise, diffusion_steps=7, embedding_scale=1))
     if text.strip() == "":
         raise gr.Error("You must enter some text")