[Cherry-picked 2.0.1] Fix path-like object support in FFmpeg dispatcher

test/torchaudio_unittest/backend/dispatcher/ffmpeg/info_test.py

@@ -1,6 +1,7 @@
 import io
 import itertools
 import os
+import pathlib
 import tarfile
 from contextlib import contextmanager
 from functools import partial
@@ -35,6 +36,24 @@
 class TestInfo(TempDirMixin, PytorchTestCase):
     _info = partial(get_info_func(), backend="ffmpeg")
 
+    def test_pathlike(self):
+        """FFmpeg dispatcher can query audio data from pathlike object"""
+        sample_rate = 16000
+        dtype = "float32"
+        num_channels = 2
+        duration = 1
+
+        path = self.get_temp_path("data.wav")
+        data = get_wav_data(dtype, num_channels, normalize=False, num_frames=duration * sample_rate)
+        save_wav(path, data, sample_rate)
+
+        info = self._info(pathlib.Path(path))
+        assert info.sample_rate == sample_rate
+        assert info.num_frames == sample_rate * duration
+        assert info.num_channels == num_channels
+        assert info.bits_per_sample == sox_utils.get_bit_depth(dtype)
+        assert info.encoding == get_encoding("wav", dtype)
+
     @parameterized.expand(
         list(
             itertools.product(

test/torchaudio_unittest/backend/dispatcher/ffmpeg/load_test.py

@@ -1,5 +1,6 @@
 import io
 import itertools
+import pathlib
 import tarfile
 from functools import partial
 
@@ -125,6 +126,21 @@ def assert_wav(self, dtype, sample_rate, num_channels, normalize, duration):
 class TestLoad(LoadTestBase):
     """Test the correctness of `self._load` for various formats"""
 
+    def test_pathlike(self):
+        """FFmpeg dispatcher can load waveform from pathlike object"""
+        sample_rate = 16000
+        dtype = "float32"
+        num_channels = 2
+        duration = 1
+
+        path = self.get_temp_path("data.wav")
+        data = get_wav_data(dtype, num_channels, normalize=False, num_frames=duration * sample_rate)
+        save_wav(path, data, sample_rate)
+
+        waveform, sr = self._load(pathlib.Path(path))
+        self.assertEqual(sr, sample_rate)
+        self.assertEqual(waveform, data)
+
     @parameterized.expand(
         list(
             itertools.product(

test/torchaudio_unittest/backend/dispatcher/ffmpeg/save_test.py

@@ -1,5 +1,6 @@
 import io
 import os
+import pathlib
 import subprocess
 import sys
 from functools import partial
@@ -146,6 +147,17 @@ def assert_save_consistency(
 @skipIfNoExec("ffmpeg")
 @skipIfNoFFmpeg
 class SaveTest(SaveTestBase):
+    def test_pathlike(self):
+        """FFmpeg dispatcher can save audio data to pathlike object"""
+        sample_rate = 16000
+        dtype = "float32"
+        num_channels = 2
+        duration = 1
+
+        path = self.get_temp_path("data.wav")
+        data = get_wav_data(dtype, num_channels, normalize=False, num_frames=duration * sample_rate)
+        self._save(pathlib.Path(path), data, sample_rate)
+
     @nested_params(
         ["path", "fileobj", "bytesio"],
         [

torchaudio/_backend/utils.py

@@ -82,7 +82,7 @@ def info(uri: Union[BinaryIO, str, os.PathLike], format: Optional[str], buffer_s
         if hasattr(uri, "read"):
             metadata = info_audio_fileobj(uri, format, buffer_size=buffer_size)
         else:
-            metadata = info_audio(uri, format)
+            metadata = info_audio(os.path.normpath(uri), format)
         metadata.bits_per_sample = _get_bits_per_sample(metadata.encoding, metadata.bits_per_sample)
         metadata.encoding = _map_encoding(metadata.encoding)
         return metadata
@@ -108,7 +108,7 @@ def load(
                 buffer_size,
             )
         else:
-            return load_audio(uri, frame_offset, num_frames, normalize, channels_first, format)
+            return load_audio(os.path.normpath(uri), frame_offset, num_frames, normalize, channels_first, format)
 
     @staticmethod
     def save(
@@ -122,7 +122,7 @@ def save(
         buffer_size: int = 4096,
     ) -> None:
         save_audio(
-            uri,
+            os.path.normpath(uri),
             src,
             sample_rate,
             channels_first,