fix: audio device for different torchaudio version

utils.py

@@ -1912,7 +1912,7 @@ 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) 
-                if torchaudio.__version__ == "2.8.0":
+                if "2.8.0" in torchaudio.__version__:
                     audio_device = audio.device
                     audio = audio.cpu()
                     audio = _resample_buffer[sample_rate](audio)