Fix padding

inference.py

@@ -26,7 +26,7 @@ def decode_tts(tokens, quantizer, n_codebooks, n_original_tokens, start_audio_to
     if reminder:
         # pad if last frame is incomplete
         pad_tokens = torch.zeros(n_codebooks - reminder, device="cuda")
-        audio_tokens = torch.cat([audio_tokens, pad_tokens[reminder:n_codebooks]], dim=0)
+        audio_tokens = torch.cat([audio_tokens, pad_tokens], dim=0)
 
     transposed = audio_tokens.view(-1, n_codebooks).t()
     codes = transposed.view(n_codebooks, 1, -1).to(device)