Refactor compat (#3518)

Summary:
The I/O functions in _compat module was introduced there so that
everything related to FFmpeg is in torchaudio.io and FFmpeg library
initialization can be carried out in `torchaudio.io.__init__`.

Now that this constraint is removed, (all the initialization happens
at `torchaudio._extension.__init__`) and `_compat` is only used by
FFmpeg dispatcher backend, we move the module to `torchaudio._backend`
for better locality.

Pull Request resolved: https://github.com/pytorch/audio/pull/3518

Reviewed By: huangruizhe

Differential Revision: D47877412

Pulled By: mthrok

fbshipit-source-id: aa18c8cb6e5d5360950df5158c33c653e37c565f

examples/tutorials/device_avsr.py

@@ -42,13 +42,12 @@ attention due to its robustness against noise.
 
 """
 
+import numpy as np
+import sentencepiece as spm
 import torch
 import torchaudio
 import torchvision
 
-import numpy as np
-import sentencepiece as spm
-
 ######################################################################
 # Overview
 # --------

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

@@ -5,9 +5,9 @@ import tarfile
 from functools import partial
 
 from parameterized import parameterized
+from torchaudio._backend.ffmpeg import _parse_save_args
 from torchaudio._backend.utils import get_load_func
 from torchaudio._internal import module_utils as _mod_utils
-from torchaudio.io._compat import _parse_save_args
 
 from torchaudio_unittest.backend.dispatcher.sox.common import name_func
 from torchaudio_unittest.common_utils import (

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

@@ -7,8 +7,8 @@ from functools import partial
 
 import torch
 from parameterized import parameterized
+from torchaudio._backend.ffmpeg import _parse_save_args
 from torchaudio._backend.utils import get_save_func
-from torchaudio.io._compat import _parse_save_args
 
 from torchaudio_unittest.backend.dispatcher.sox.common import get_enc_params, name_func
 from torchaudio_unittest.common_utils import (

torchaudio/io/_compat.py → torchaudio/_backend/ffmpeg.py

@@ -9,9 +9,10 @@ from torchaudio.io import StreamWriter
 
 if torchaudio._extension._FFMPEG_EXT is not None:
     StreamReaderFileObj = torchaudio._extension._FFMPEG_EXT.StreamReaderFileObj
+else:
+    StreamReaderFileObj = object
 
 
-# Note: need to comply TorchScript syntax -- need annotation and no f-string nor global
 def info_audio(
     src: str,
     format: Optional[str],
@@ -241,7 +242,6 @@ def _parse_save_args(
     return muxer, encoder, sample_fmt
 
 
-# NOTE: in contrast to load_audio* and info_audio*, this function is NOT compatible with TorchScript.
 def save_audio(
     uri: Union[BinaryIO, str, os.PathLike],
     src: torch.Tensor,

torchaudio/_backend/utils.py

@@ -5,12 +5,12 @@ from functools import lru_cache
 from typing import BinaryIO, Dict, Optional, Tuple, Union
 
 import torch
-import torchaudio.backend.soundfile_backend as soundfile_backend
+
 from torchaudio._extension import _FFMPEG_EXT, _SOX_INITIALIZED
+from torchaudio.backend import soundfile_backend
 from torchaudio.backend.common import AudioMetaData
 
-if _FFMPEG_EXT is not None:
-    from torchaudio.io._compat import info_audio, info_audio_fileobj, load_audio, load_audio_fileobj, save_audio
+from . import ffmpeg
 
 
 class Backend(ABC):
@@ -80,9 +80,9 @@ class FFmpegBackend(Backend):
     @staticmethod
     def info(uri: Union[BinaryIO, str, os.PathLike], format: Optional[str], buffer_size: int = 4096) -> AudioMetaData:
         if hasattr(uri, "read"):
-            metadata = info_audio_fileobj(uri, format, buffer_size=buffer_size)
+            metadata = ffmpeg.info_audio_fileobj(uri, format, buffer_size=buffer_size)
         else:
-            metadata = info_audio(os.path.normpath(uri), format)
+            metadata = ffmpeg.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
@@ -98,7 +98,7 @@ class FFmpegBackend(Backend):
         buffer_size: int = 4096,
     ) -> Tuple[torch.Tensor, int]:
         if hasattr(uri, "read"):
-            return load_audio_fileobj(
+            return ffmpeg.load_audio_fileobj(
                 uri,
                 frame_offset,
                 num_frames,
@@ -108,7 +108,7 @@ class FFmpegBackend(Backend):
                 buffer_size,
             )
         else:
-            return load_audio(os.path.normpath(uri), frame_offset, num_frames, normalize, channels_first, format)
+            return ffmpeg.load_audio(os.path.normpath(uri), frame_offset, num_frames, normalize, channels_first, format)
 
     @staticmethod
     def save(
@@ -121,7 +121,7 @@ class FFmpegBackend(Backend):
         bits_per_sample: Optional[int] = None,
         buffer_size: int = 4096,
     ) -> None:
-        save_audio(
+        ffmpeg.save_audio(
             uri,
             src,
             sample_rate,