Replace resampy with torchaudio

Signed-off-by: Jael Gu <mengjia.gu@zilliz.com>

requirements.txt

@@ -1,6 +1,5 @@
 torch>=1.9.0
-numpy>=1.19.5
-resampy
+torchaudio>=0.9.0
 
 towhee>=0.7.0
 towhee.models

vggish.py

@@ -87,10 +87,15 @@ class Vggish(NNOperator):
         assert dtype.kind == 'f'
 
         if wav.dtype.kind in 'iu':
-            ii = numpy.iinfo(wav.dtype)
-            abs_max = 2 ** (ii.bits - 1)
-            offset = ii.min + abs_max
-            return (wav.astype(dtype) - offset) / abs_max
+            # ii = numpy.iinfo(wav.dtype)
+            # abs_max = 2 ** (ii.bits - 1)
+            # offset = ii.min + abs_max
+            # return (wav.astype(dtype) - offset) / abs_max
+            if wav.dtype != 'int16':
+                wav = (wav >> 16).astype(numpy.int16)
+            assert wav.dtype == 'int16'
+            wav = (wav / 32768.0).astype(dtype)
+            return wav
         else:
             log.warning('Converting float dtype from %s to %s.', wav.dtype, dtype)
             return wav.astype(dtype)

vggish_input.py

@@ -17,8 +17,8 @@
 
 # Modification: Return torch tensors rather than numpy arrays
 import torch
+import torchaudio
 import numpy as np
-import resampy
 
 import mel_features
 import vggish_params
@@ -46,7 +46,9 @@ def waveform_to_examples(data, sample_rate, return_tensor=True):
         data = np.mean(data, axis=1)
     # Resample to the rate assumed by VGGish.
     if sample_rate != vggish_params.SAMPLE_RATE:
-        data = resampy.resample(data, sample_rate, vggish_params.SAMPLE_RATE)
+        data = torch.from_numpy(data)
+        resampler = torchaudio.transforms.Resample(sample_rate, vggish_params.SAMPLE_RATE, dtype=data.dtype)
+        data = resampler(data).cpu().detach().numpy()
 
     # Compute log mel spectrogram features.
     log_mel = mel_features.log_mel_spectrogram(