Set individual backends

test/test_batch_consistency.py

@@ -47,6 +47,7 @@ def test_griffinlim(self):
         )
 
     def test_detect_pitch_frequency(self):
+        common_utils.set_audio_backend('default')
         filenames = [
             'steam-train-whistle-daniel_simon.wav',  # 2ch 44100Hz
             # Files from https://www.mediacollege.com/audio/tone/download/
@@ -79,6 +80,7 @@ def test_overdrive(self):
         self.assert_batch_consistencies(F.overdrive, waveform, gain=45, colour=30)
 
     def test_phaser(self):
+        common_utils.set_audio_backend('default')
         filepath = common_utils.get_asset_path("whitenoise.wav")
         waveform, sample_rate = torchaudio.load(filepath)
         self.assert_batch_consistencies(F.phaser, waveform, sample_rate)
@@ -97,6 +99,7 @@ def test_sliding_window_cmn(self):
         self.assert_batch_consistencies(F.sliding_window_cmn, waveform, center=False, norm_vars=False)
 
     def test_vad(self):
+        common_utils.set_audio_backend('default')
         filepath = common_utils.get_asset_path("vad-go-mono-32000.wav")
         waveform, sample_rate = torchaudio.load(filepath)
         self.assert_batch_consistencies(F.vad, waveform, sample_rate=sample_rate)
@@ -168,6 +171,7 @@ def test_batch_compute_deltas(self):
         self.assertEqual(computed, expected)
 
     def test_batch_mulaw(self):
+        common_utils.set_audio_backend('default')
         test_filepath = common_utils.get_asset_path('steam-train-whistle-daniel_simon.wav')
         waveform, _ = torchaudio.load(test_filepath)  # (2, 278756), 44100
 
@@ -193,6 +197,7 @@ def test_batch_mulaw(self):
         self.assertEqual(computed, expected)
 
     def test_batch_spectrogram(self):
+        common_utils.set_audio_backend('default')
         test_filepath = common_utils.get_asset_path('steam-train-whistle-daniel_simon.wav')
         waveform, _ = torchaudio.load(test_filepath)  # (2, 278756), 44100
 
@@ -204,6 +209,7 @@ def test_batch_spectrogram(self):
         self.assertEqual(computed, expected)
 
     def test_batch_melspectrogram(self):
+        common_utils.set_audio_backend('default')
         test_filepath = common_utils.get_asset_path('steam-train-whistle-daniel_simon.wav')
         waveform, _ = torchaudio.load(test_filepath)  # (2, 278756), 44100
 
@@ -215,6 +221,7 @@ def test_batch_melspectrogram(self):
         self.assertEqual(computed, expected)
 
     def test_batch_mfcc(self):
+        common_utils.set_audio_backend('default')
         test_filepath = common_utils.get_asset_path('steam-train-whistle-daniel_simon.wav')
         waveform, _ = torchaudio.load(test_filepath)
 
@@ -226,6 +233,7 @@ def test_batch_mfcc(self):
         self.assertEqual(computed, expected, atol=1e-4, rtol=1e-5)
 
     def test_batch_TimeStretch(self):
+        common_utils.set_audio_backend('default')
         test_filepath = common_utils.get_asset_path('steam-train-whistle-daniel_simon.wav')
         waveform, _ = torchaudio.load(test_filepath)  # (2, 278756), 44100
 
@@ -260,6 +268,7 @@ def test_batch_TimeStretch(self):
         self.assertEqual(computed, expected, atol=1e-5, rtol=1e-5)
 
     def test_batch_Fade(self):
+        common_utils.set_audio_backend('default')
         test_filepath = common_utils.get_asset_path('steam-train-whistle-daniel_simon.wav')
         waveform, _ = torchaudio.load(test_filepath)  # (2, 278756), 44100
         fade_in_len = 3000
@@ -273,6 +282,7 @@ def test_batch_Fade(self):
         self.assertEqual(computed, expected)
 
     def test_batch_Vol(self):
+        common_utils.set_audio_backend('default')
         test_filepath = common_utils.get_asset_path('steam-train-whistle-daniel_simon.wav')
         waveform, _ = torchaudio.load(test_filepath)  # (2, 278756), 44100
 

test/test_librosa_compatibility.py

@@ -159,6 +159,7 @@ def _load_audio_asset(*asset_paths, **kwargs):
 class TestTransforms(common_utils.TorchaudioTestCase):
     """Test suite for functions in `transforms` module."""
     def assert_compatibilities(self, n_fft, hop_length, power, n_mels, n_mfcc, sample_rate):
+        common_utils.set_audio_backend('default')
         sound, sample_rate = _load_audio_asset('sinewave.wav')
         sound_librosa = sound.cpu().numpy().squeeze()  # (64000)
 
@@ -275,6 +276,7 @@ def test_MelScale(self):
         hop_length = n_fft // 4
 
         # Prepare spectrogram input. We use torchaudio to compute one.
+        common_utils.set_audio_backend('default')
         sound, sample_rate = _load_audio_asset('whitenoise_1min.mp3')
         sound = sound.mean(dim=0, keepdim=True)
         spec_ta = F.spectrogram(
@@ -297,6 +299,7 @@ def test_InverseMelScale(self):
         hop_length = n_fft // 4
 
         # Prepare mel spectrogram input. We use torchaudio to compute one.
+        common_utils.set_audio_backend('default')
         sound, sample_rate = _load_audio_asset(
             'steam-train-whistle-daniel_simon.wav', offset=2**10, num_frames=2**14)
         sound = sound.mean(dim=0, keepdim=True)