Update app.py

app.py

@@ -65,6 +65,7 @@ for v in voicelist:
 #     # return (24000, styletts2importable.inference(text, voices[v], alpha=0.3, beta=0.7, diffusion_steps=7, embedding_scale=1))
 #     return (24000, styletts2importable.inference(text, voices[v], alpha=0.3, beta=0.7, diffusion_steps=multispeakersteps, embedding_scale=1))
 if not torch.cuda.is_available(): INTROTXT += "\n\n### on CPU, it'll run rather slower, but not too much."
+if not torch.cuda.is_available(): device = "cpu" else "cuda"
 @spaces.GPU
 def synthesize(text, voice, lngsteps,embscale,alpha, beta, password, progress=gr.Progress()):
     if text.strip() == "":
@@ -79,7 +80,7 @@ def synthesize(text, voice, lngsteps,embscale,alpha, beta, password, progress=gr
     audios = []
     for t in progress.tqdm(texts):
         print(t)
-        audios.append(styletts2importable.inference(t, voices[v], alpha=alpha, beta=beta, diffusion_steps=lngsteps, embedding_scale=embscale))
+        audios.append(styletts2importable.inference(t, voices[v].to(device), alpha=alpha, beta=beta, diffusion_steps=lngsteps, embedding_scale=embscale))
     return (24000, np.concatenate(audios))
 # def longsynthesize(text, voice, lngsteps, password, progress=gr.Progress()):
 #     if password == os.environ['ACCESS_CODE']:
@@ -97,7 +98,8 @@ def synthesize(text, voice, lngsteps,embscale,alpha, beta, password, progress=gr
 #         return (24000, np.concatenate(audios))
 #     else:
 #         raise gr.Error('Wrong access code')
-    
+
+@spaces.GPU
 def clsynthesize(text, voice, vcsteps, embscale, alpha, beta, progress=gr.Progress()):