Update utils.py

utils.py

@@ -1909,6 +1909,10 @@ def extract_speech_token(model, feature_extractor, utts, batch_size=128):
                         orig_freq=sample_rate,
                         new_freq=16000
                     ).to(device)
+                    # 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 = _resample_buffer[sample_rate](audio)
             # if audio.shape[0] > 1:
             #     audio = audio[:1]