fix(torchaudio): use `soundfile` as torchaudio backend

inference/infer_tool.py

@@ -267,6 +267,7 @@ class Svc(object):
               second_encoding = False,
               loudness_envelope_adjustment = 1
               ):
+        torchaudio.set_audio_backend("soundfile")
         wav, sr = torchaudio.load(raw_path)
         if not hasattr(self,"audio_resample_transform") or self.audio16k_resample_transform.orig_freq != sr:
             self.audio_resample_transform = torchaudio.transforms.Resample(sr,self.target_sample)