Merge pull request #19 from thunn/thunn/update_load_to_convert_stereo…

…_to_mono

update load fn to convert stereo to mono

penn/load.py

@@ -1,4 +1,5 @@
 import json
+import warnings
 
 import torchaudio
 
@@ -9,6 +10,11 @@ def audio(file):
     """Load audio from disk"""
     audio, sample_rate = torchaudio.load(file)
 
+    # if audio is stereo convert to mono
+    if audio.size(0) == 2:
+        warnings.warn(f'Converting stereo audio to mono: {file}')
+        audio = audio.mean(dim=0, keepdim=True)
+
     # Maybe resample
     return penn.resample(audio, sample_rate)
 

test/conftest.py

@@ -14,3 +14,8 @@
 def audio():
     """Retrieve the test audio"""
     return penn.load.audio(Path(__file__).parent / 'assets' / 'gershwin.wav')
+
+@pytest.fixture(scope='session')
+def audio_stereo():
+    """Retrieve the test audio"""
+    return penn.load.audio(Path(__file__).parent / 'assets' / '500Hz_stereo.wav')

test/test_core.py

@@ -14,3 +14,12 @@ def test_infer(audio):
         center='half-hop')
     shape = (1, audio.shape[1] // penn.HOPSIZE)
     assert pitch.shape == periodicity.shape == shape
+
+def test_infer_stereo(audio_stereo):
+    """Test that inference on stereo audio produces the correct shape"""
+    pitch, periodicity = penn.from_audio(
+        audio_stereo,
+        penn.SAMPLE_RATE,
+        center='half-hop')
+    shape = (1, audio_stereo.shape[1] // penn.HOPSIZE)
+    assert pitch.shape == periodicity.shape == shape