[Cherry-picks for 2.8] (#3976)

Author: NicolasHug

.github/scripts/unittest-linux/install.sh

@@ -75,6 +75,7 @@ case $GPU_ARCH_TYPE in
 esac
 PYTORCH_WHEEL_INDEX="https://download.pytorch.org/whl/${UPLOAD_CHANNEL}/${GPU_ARCH_ID}"
 pip install --progress-bar=off --pre torch --index-url="${PYTORCH_WHEEL_INDEX}"
+pip install --progress-bar=off --pre torchcodec --index-url="https://download.pytorch.org/whl/nightly/cpu"
 
 
 # 2. Install torchaudio
@@ -86,6 +87,10 @@ python setup.py install
 
 # 3. Install Test tools
 printf "* Installing test tools\n"
+# On this CI, for whatever reason, we're only able to install ffmpeg 4.
+conda install -y "ffmpeg<5"
+python -c "import torch; import torchaudio; import torchcodec; print(torch.__version__, torchaudio.__version__, torchcodec.__version__)"
+
 NUMBA_DEV_CHANNEL=""
 if [[ "$(python --version)" = *3.9* || "$(python --version)" = *3.10* ]]; then
     # Numba isn't available for Python 3.9 and 3.10 except on the numba dev channel and building from source fails
@@ -94,7 +99,7 @@ if [[ "$(python --version)" = *3.9* || "$(python --version)" = *3.10* ]]; then
 fi
 (
     set -x
-    conda install -y -c conda-forge ${NUMBA_DEV_CHANNEL} sox libvorbis parameterized 'requests>=2.20' 'ffmpeg>=6,<7'
+    conda install -y -c conda-forge ${NUMBA_DEV_CHANNEL} sox libvorbis parameterized 'requests>=2.20'
     pip install kaldi-io SoundFile librosa coverage pytest pytest-cov scipy expecttest unidecode inflect Pillow sentencepiece pytorch-lightning 'protobuf<4.21.0' demucs tinytag pyroomacoustics flashlight-text git+https://github.com/kpu/kenlm
 
     # TODO: might be better to fix the single call to `pip install` above

docs/source/torchaudio.rst

@@ -7,9 +7,13 @@ torchaudio
     Starting with version 2.8, we are refactoring TorchAudio to transition it
     into a maintenance phase. As a result:
 
-    - The APIs listed below are deprecated in 2.8 and will be removed in 2.9.
+    - Most APIs listed below are deprecated in 2.8 and will be removed in 2.9.
     - The decoding and encoding capabilities of PyTorch for both audio and video
-      are being consolidated into TorchCodec.
+      are being consolidated into TorchCodec. For convenience, we provide
+      :func:`~torchaudio.load_with_torchcodec` as a replacement for
+      :func:`~torchaudio.load` and :func:`~torchaudio.save_with_torchcodec` as a
+      replacement for :func:`~torchaudio.save`, but we recommend that you port
+      your code to native torchcodec APIs.
 
     Please see https://github.com/pytorch/audio/issues/3902 for more information.
 
@@ -26,7 +30,9 @@ it easy to handle audio data.
 
    info
    load
+   load_with_torchcodec
    save
+   save_with_torchcodec
    list_audio_backends
 
 .. _backend:

src/torchaudio/__init__.py

@@ -7,17 +7,16 @@
     get_audio_backend as _get_audio_backend,
     info as _info,
     list_audio_backends as _list_audio_backends,
-    load as _load,
-    save as _save,
+    load,
+    save,
     set_audio_backend as _set_audio_backend,
 )
+from ._torchcodec import load_with_torchcodec, save_with_torchcodec
 
 AudioMetaData = dropping_class_io_support(_AudioMetaData)
 get_audio_backend = dropping_io_support(_get_audio_backend)
 info = dropping_io_support(_info)
 list_audio_backends = dropping_io_support(_list_audio_backends)
-load = dropping_io_support(_load)
-save = dropping_io_support(_save)
 set_audio_backend = dropping_io_support(_set_audio_backend)
 
 from . import (  # noqa: F401
@@ -45,6 +44,8 @@
 __all__ = [
     "AudioMetaData",
     "load",
+    "load_with_torchcodec",
+    "save_with_torchcodec",
     "info",
     "save",
     "io",

src/torchaudio/_backend/utils.py

@@ -1,6 +1,7 @@
 import os
 from functools import lru_cache
 from typing import BinaryIO, Dict, Optional, Tuple, Type, Union
+import warnings
 
 import torch
 
@@ -127,6 +128,14 @@ def load(
     ) -> Tuple[torch.Tensor, int]:
         """Load audio data from source.
 
+        .. warning::
+            In 2.9, this function's implementation will be changed to use
+            :func:`~torchaudio.load_with_torchcodec` under the hood. Some
+            parameters like ``normalize``, ``format``, ``buffer_size``, and
+            ``backend`` will be ignored. We recommend that you port your code to
+            rely directly on TorchCodec's decoder instead:
+            https://docs.pytorch.org/torchcodec/stable/generated/torchcodec.decoders.AudioDecoder.html#torchcodec.decoders.AudioDecoder.
+
         By default (``normalize=True``, ``channels_first=True``), this function returns Tensor with
         ``float32`` dtype, and the shape of `[channel, time]`.
 
@@ -201,6 +210,14 @@ def load(
                 integer type, else ``float32`` type. If ``channels_first=True``, it has
                 `[channel, time]` else `[time, channel]`.
         """
+        warnings.warn(
+            "In 2.9, this function's implementation will be changed to use "
+            "torchaudio.load_with_torchcodec` under the hood. Some "
+            "parameters like ``normalize``, ``format``, ``buffer_size``, and "
+            "``backend`` will be ignored. We recommend that you port your code to "
+            "rely directly on TorchCodec's decoder instead: "
+            "https://docs.pytorch.org/torchcodec/stable/generated/torchcodec.decoders.AudioDecoder.html#torchcodec.decoders.AudioDecoder."
+        )
         backend = dispatcher(uri, format, backend)
         return backend.load(uri, frame_offset, num_frames, normalize, channels_first, format, buffer_size)
 
@@ -235,6 +252,14 @@ def save(
     ):
         """Save audio data to file.
 
+        .. warning::
+            In 2.9, this function's implementation will be changed to use
+            :func:`~torchaudio.save_with_torchcodec` under the hood. Some
+            parameters like format, encoding, bits_per_sample, buffer_size, and
+            ``backend`` will be ignored. We recommend that you port your code to
+            rely directly on TorchCodec's decoder instead:
+            https://docs.pytorch.org/torchcodec/stable/generated/torchcodec.encoders.AudioEncoder
+
         Note:
             The formats this function can handle depend on the availability of backends.
             Please use the following functions to fetch the supported formats.
@@ -309,6 +334,14 @@ def save(
                 Refer to http://sox.sourceforge.net/soxformat.html for more details.
 
         """
+        warnings.warn(
+            "In 2.9, this function's implementation will be changed to use "
+            "torchaudio.save_with_torchcodec` under the hood. Some "
+            "parameters like format, encoding, bits_per_sample, buffer_size, and "
+            "``backend`` will be ignored. We recommend that you port your code to "
+            "rely directly on TorchCodec's encoder instead: "
+            "https://docs.pytorch.org/torchcodec/stable/generated/torchcodec.encoders.AudioEncoder"
+        )
         backend = dispatcher(uri, format, backend)
         return backend.save(
             uri, src, sample_rate, channels_first, format, encoding, bits_per_sample, buffer_size, compression