Update app.py

app.py

@@ -94,11 +94,11 @@ def build_prompt(text: str, voice: str):
 
 def decode_block(block7: list[int]) -> bytes:
     l1,l2,l3=[],[],[]
-    l1.append(block7[0] - AUDIO_BASE) # Subtract AUDIO_BASE
-    l2.append((block7[1] - AUDIO_BASE) - 4096) # Subtract AUDIO_BASE then position offset
-    l3 += [(block7[2] - AUDIO_BASE) - 8192, (block7[3] - AUDIO_BASE) - 12288] # Subtract AUDIO_BASE then position offsets
-    l2.append((block7[4] - AUDIO_BASE) - 16384) # Subtract AUDIO_BASE then position offset
-    l3 += [(block7[5] - AUDIO_BASE) - 20480, (block7[6] - AUDIO_BASE) - 24576] # Subtract AUDIO_BASE then position offsets
+    l1.append((block7[0] - AUDIO_BASE) - 0 * 4096) # Subtract AUDIO_BASE and position 0 offset
+    l2.append((block7[1] - AUDIO_BASE) - 1 * 4096) # Subtract AUDIO_BASE and position 1 offset
+    l3 += [(block7[2] - AUDIO_BASE) - 2 * 4096, (block7[3] - AUDIO_BASE) - 3 * 4096] # Subtract AUDIO_BASE and position offsets
+    l2.append((block7[4] - AUDIO_BASE) - 4 * 4096) # Subtract AUDIO_BASE and position 4 offset
+    l3 += [(block7[5] - AUDIO_BASE) - 5 * 4096, (block7[6] - AUDIO_BASE) - 6 * 4096] # Subtract AUDIO_BASE and position offsets
 
     with torch.no_grad():
         codes = [torch.tensor(x, device=device).unsqueeze(0)