Update app.py

app.py

@@ -75,7 +75,6 @@ def synthesize_speech(text, embedding_path):
         # Synthesize each chunk
         audio_chunks = []
         for chunk in text_chunks:
-            start_time = time.time()
             out = tts.synthesizer.tts_model.inference(
                 chunk,
                 "ru",
@@ -85,10 +84,13 @@ def synthesize_speech(text, embedding_path):
                 length_penalty=1.0,
                 repetition_penalty=2.0,
             )
-            # Convert tensor to numpy array properly
-            audio = out["wav"].squeeze().cpu().numpy()
-            audio_chunks.append(audio)
-            
+            # Handle both tensor and numpy array outputs
+            wav = out["wav"].squeeze()
+            if isinstance(wav, torch.Tensor):
+                audio_chunks.append(wav.cpu().numpy())
+            else:
+                audio_chunks.append(wav)
+        
         # Combine and save audio
         full_audio = np.concatenate(audio_chunks)
         output_path = "output.wav"