Update utils.py

utils.py

@@ -1912,8 +1912,10 @@ def extract_speech_token(model, feature_extractor, utts, batch_size=128):
                     # make sure kernel also on device
                     if hasattr(_resample_buffer[sample_rate], "kernel"):
                         _resample_buffer[sample_rate].kernel = _resample_buffer[sample_rate].kernel.to(device)
-
+                        
+                audio = audio.cpu()
                 audio = _resample_buffer[sample_rate](audio)
+                audio = audio.to(device)
             # if audio.shape[0] > 1:
             #     audio = audio[:1]
             audio = audio[0]