Skip to content

GitLab

Unverified Commit 33dc817c authored by Prabhat Roy's avatar Prabhat Roy Committed by GitHub
Browse files

Remove legacy backends (#1311)

parent 086467a1
Expand all Show whitespace changes
Inline Side-by-side
Showing with 432 additions and 1829 deletions
docs/source/backend.rst
View file @ 33dc817c
...@@ -11,11 +11,7 @@ Overview ...@@ -11,11 +11,7 @@ Overview
There are currently four implementations available. There are currently four implementations available.
* :ref:`"sox_io" <sox_io_backend>` (default on Linux/macOS) * :ref:`"sox_io" <sox_io_backend>` (default on Linux/macOS)
* :ref:`"sox" <sox_backend>` (deprecated, will be removed in 0.9.0 release)
* :ref:`"soundfile" <soundfile_backend>` (default on Windows) * :ref:`"soundfile" <soundfile_backend>` (default on Windows)
* :ref:`"soundfile" (legacy interface) <soundfile_legacy_backend>` (deprecated, will be removed in 0.9.0 release)
The use of ``"sox"`` backend is strongly discouraged as it cannot correctly handle formats other than 16-bit integer WAV. See `#726 <https://github.com/pytorch/audio/pull/726>`_ for the detail.
.. note:: .. note::
Instead of calling functions in ``torchaudio.backend`` directly, please use ``torchaudio.info``, ``torchaudio.load``, ``torchaudio.load_wav`` and ``torchaudio.save`` with proper backend set with :func:`torchaudio.set_audio_backend`. Instead of calling functions in ``torchaudio.backend`` directly, please use ``torchaudio.info``, ``torchaudio.load``, ``torchaudio.load_wav`` and ``torchaudio.save`` with proper backend set with :func:`torchaudio.set_audio_backend`.
...@@ -23,32 +19,10 @@ The use of ``"sox"`` backend is strongly discouraged as it cannot correctly hand ...@@ -23,32 +19,10 @@ The use of ``"sox"`` backend is strongly discouraged as it cannot correctly hand
Availability Availability
------------ ------------
``"sox"`` and ``"sox_io"`` backends require C++ extension module, which is included in Linux/macOS binary distributions. These backends are not available on Windows. ``"sox_io"`` backend requires C++ extension module, which is included in Linux/macOS binary distributions. This backend is not available on Windows.
``"soundfile"`` backend requires ``SoundFile``. Please refer to `the SoundFile documentation <https://pysoundfile.readthedocs.io/en/latest/>`_ for the installation. ``"soundfile"`` backend requires ``SoundFile``. Please refer to `the SoundFile documentation <https://pysoundfile.readthedocs.io/en/latest/>`_ for the installation.
Changes in default backend and deprecation
------------------------------------------
Backend module is going through a major overhaul. The following table summarizes the timeline for the deprecations and removals.
+--------------------+-----------------------+------------------------+
| **Backend** | **0.8.0** | **0.9.0** |
+====================+=======================+========================+
| ``"sox_io"`` | Default on Linx/macOS | Default on Linux/macOS |
+--------------------+-----------------------+------------------------+
| ``"sox"`` | Available | Removed |
| (deprecated) | | |
+--------------------+-----------------------+------------------------+
| ``"soundfile"`` | Default on Windows | Default on Windows |
+--------------------+-----------------------+------------------------+
| ``"soundfile"`` | Available | Removed |
| (legacy interface, | | |
| deprecated) | | |
+--------------------+-----------------------+------------------------+
* The ``"sox"`` and ``"soundfile" (legacy interface)`` backends are deprecated and will be removed in 0.9.0 release.
Common Data Structure Common Data Structure
~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~
...@@ -59,16 +33,6 @@ AudioMetaData ...@@ -59,16 +33,6 @@ AudioMetaData
.. autoclass:: torchaudio.backend.common.AudioMetaData .. autoclass:: torchaudio.backend.common.AudioMetaData
SignalInfo (Deprecated)
-----------------------
.. autoclass:: torchaudio.backend.common.SignalInfo
EncodingInfo (Deprecated)
-------------------------
.. autoclass:: torchaudio.backend.common.EncodingInfo
.. _sox_io_backend: .. _sox_io_backend:
Sox IO Backend Sox IO Backend
...@@ -102,44 +66,6 @@ save ...@@ -102,44 +66,6 @@ save
.. autofunction:: torchaudio.backend.sox_io_backend.save .. autofunction:: torchaudio.backend.sox_io_backend.save
.. _sox_backend:
Sox Backend (Deprecated)
~~~~~~~~~~~~~~~~~~~~~~~~
The ``"sox"`` backend is available on Linux/macOS and not available on Windows. This backend is deprecated and will be removed in ``0.9.0`` release.
You can switch from another backend to ``sox`` backend with the following;
.. code::
torchaudio.set_audio_backend("sox")
info
----
.. autofunction:: torchaudio.backend.sox_backend.info
load
----
.. autofunction:: torchaudio.backend.sox_backend.load
.. autofunction:: torchaudio.backend.sox_backend.load_wav
save
----
.. autofunction:: torchaudio.backend.sox_backend.save
others
------
.. automodule:: torchaudio.backend.sox_backend
:members:
:exclude-members: info, load, load_wav, save
.. _soundfile_backend: .. _soundfile_backend:
Soundfile Backend Soundfile Backend
...@@ -153,48 +79,13 @@ You can switch from another backend to the ``"soundfile"`` backend with the foll ...@@ -153,48 +79,13 @@ You can switch from another backend to the ``"soundfile"`` backend with the foll
torchaudio.set_audio_backend("soundfile") torchaudio.set_audio_backend("soundfile")
.. note::
If you are switching from `"soundfile" (legacy interface) <soundfile_legacy_backend>` backend, set ``torchaudio.USE_SOUNDFILE_LEGACY_INTERFACE`` flag **before** switching the backend.
info info
---- ----
.. autofunction:: torchaudio.backend._soundfile_backend.info
load
----
.. autofunction:: torchaudio.backend._soundfile_backend.load
.. autofunction:: torchaudio.backend._soundfile_backend.load_wav
save
----
.. autofunction:: torchaudio.backend._soundfile_backend.save
.. _soundfile_legacy_backend:
Legacy Interface (Deprecated)
-----------------------------
``"soundfile"`` backend with legacy interface is made available for backward compatibility reason, however this interface is deprecated and will be removed in the ``0.9.0`` release.
To switch to this backend/interface, set ``torchaudio.USE_SOUNDFILE_LEGACY_INTERFACE`` flag **before** switching the backend.
.. code::
torchaudio.USE_SOUNDFILE_LEGACY_INTERFACE = True
torchaudio.set_audio_backend("soundfile") # The legacy interface
info
^^^^
.. autofunction:: torchaudio.backend.soundfile_backend.info .. autofunction:: torchaudio.backend.soundfile_backend.info
load load
^^^^ ----
.. autofunction:: torchaudio.backend.soundfile_backend.load .. autofunction:: torchaudio.backend.soundfile_backend.load
...@@ -202,6 +93,6 @@ load ...@@ -202,6 +93,6 @@ load
save save
^^^^ ----
.. autofunction:: torchaudio.backend.soundfile_backend.save .. autofunction:: torchaudio.backend.soundfile_backend.save
test/torchaudio_unittest/backend/legacy_test.py deleted 100644 → 0
View file @ 086467a1
import os
import math
import shutil
import tempfile
import unittest
import torch
import torchaudio
from torchaudio.utils import sox_utils
from torchaudio._internal.module_utils import is_module_available
from torchaudio_unittest.common_utils import get_asset_path
BACKENDS = []
BACKENDS_MP3 = []
if is_module_available('soundfile'):
BACKENDS.append('soundfile')
if is_module_available('torchaudio._torchaudio'):
BACKENDS.append('sox')
if (
'mp3' in sox_utils.list_read_formats() and
'mp3' in sox_utils.list_write_formats()
):
BACKENDS_MP3 = ['sox']
def create_temp_assets_dir():
"""
Creates a temporary directory and moves all files from test/assets there.
Returns a Tuple[string, TemporaryDirectory] which is the folder path
and object.
"""
tmp_dir = tempfile.TemporaryDirectory()
shutil.copytree(get_asset_path(), os.path.join(tmp_dir.name, "assets"))
return tmp_dir.name, tmp_dir
class Test_LoadSave(unittest.TestCase):
test_dirpath, test_dir = create_temp_assets_dir()
test_filepath = os.path.join(test_dirpath, "assets",
"steam-train-whistle-daniel_simon.mp3")
test_filepath_wav = os.path.join(test_dirpath, "assets",
"steam-train-whistle-daniel_simon.wav")
def setUp(self):
torchaudio.USE_SOUNDFILE_LEGACY_INTERFACE = True
def test_1_save(self):
for backend in BACKENDS_MP3:
with self.subTest():
torchaudio.set_audio_backend(backend)
self._test_1_save(self.test_filepath, False)
for backend in BACKENDS:
with self.subTest():
torchaudio.set_audio_backend(backend)
self._test_1_save(self.test_filepath_wav, True)
def _test_1_save(self, test_filepath, normalization):
# load signal
x, sr = torchaudio.load(test_filepath, normalization=normalization)
# check save
new_filepath = os.path.join(self.test_dirpath, "test.wav")
torchaudio.save(new_filepath, x, sr)
self.assertTrue(os.path.isfile(new_filepath))
os.unlink(new_filepath)
# check automatic normalization
x /= 1 << 31
torchaudio.save(new_filepath, x, sr)
self.assertTrue(os.path.isfile(new_filepath))
os.unlink(new_filepath)
# test save 1d tensor
x = x[0, :] # get mono signal
x.squeeze_() # remove channel dim
torchaudio.save(new_filepath, x, sr)
self.assertTrue(os.path.isfile(new_filepath))
os.unlink(new_filepath)
# don't allow invalid sizes as inputs
with self.assertRaises(ValueError):
x.unsqueeze_(1) # L x C not C x L
torchaudio.save(new_filepath, x, sr)
with self.assertRaises(ValueError):
x.squeeze_()
x.unsqueeze_(1)
x.unsqueeze_(0) # 1 x L x 1
torchaudio.save(new_filepath, x, sr)
# don't save to folders that don't exist
with self.assertRaises(OSError):
new_filepath = os.path.join(self.test_dirpath, "no-path",
"test.wav")
torchaudio.save(new_filepath, x, sr)
def test_1_save_sine(self):
for backend in BACKENDS:
with self.subTest():
torchaudio.set_audio_backend(backend)
self._test_1_save_sine()
def _test_1_save_sine(self):
# save created file
sinewave_filepath = os.path.join(self.test_dirpath, "assets",
"sinewave.wav")
sr = 16000
freq = 440
volume = 0.3
y = (torch.cos(
2 * math.pi * torch.arange(0, 4 * sr).float() * freq / sr))
y.unsqueeze_(0)
# y is between -1 and 1, so must scale
y = (y * volume * (2**31)).long()
torchaudio.save(sinewave_filepath, y, sr)
self.assertTrue(os.path.isfile(sinewave_filepath))
# test precision
new_precision = 32
new_filepath = os.path.join(self.test_dirpath, "test.wav")
si, ei = torchaudio.info(sinewave_filepath)
torchaudio.save(new_filepath, y, sr, new_precision)
si32, ei32 = torchaudio.info(new_filepath)
self.assertEqual(si.precision, 16)
self.assertEqual(si32.precision, new_precision)
os.unlink(new_filepath)
def test_2_load(self):
for backend in BACKENDS_MP3:
with self.subTest():
torchaudio.set_audio_backend(backend)
self._test_2_load(self.test_filepath, 278756)
for backend in BACKENDS:
with self.subTest():
torchaudio.set_audio_backend(backend)
self._test_2_load(self.test_filepath_wav, 276858)
def _test_2_load(self, test_filepath, length):
# check normal loading
x, sr = torchaudio.load(test_filepath)
self.assertEqual(sr, 44100)
self.assertEqual(x.size(), (2, length))
# check offset
offset = 15
x, _ = torchaudio.load(test_filepath)
x_offset, _ = torchaudio.load(test_filepath, offset=offset)
self.assertTrue(x[:, offset:].allclose(x_offset))
# check number of frames
n = 201
x, _ = torchaudio.load(test_filepath, num_frames=n)
self.assertTrue(x.size(), (2, n))
# check channels first
x, _ = torchaudio.load(test_filepath, channels_first=False)
self.assertEqual(x.size(), (length, 2))
# check raising errors
with self.assertRaises(OSError):
torchaudio.load("file-does-not-exist.mp3")
with self.assertRaises(OSError):
tdir = os.path.join(
os.path.dirname(self.test_dirpath), "torchaudio")
torchaudio.load(tdir)
def test_2_load_nonormalization(self):
for backend in BACKENDS_MP3:
if backend == 'sox_io':
continue
with self.subTest():
torchaudio.set_audio_backend(backend)
self._test_2_load_nonormalization(self.test_filepath, 278756)
def _test_2_load_nonormalization(self, test_filepath, length):
# check no normalizing
x, _ = torchaudio.load(test_filepath, normalization=False)
self.assertTrue(x.min() <= -1.0)
self.assertTrue(x.max() >= 1.0)
# check different input tensor type
x, _ = torchaudio.load(test_filepath, torch.LongTensor(), normalization=False)
self.assertTrue(isinstance(x, torch.LongTensor))
def test_3_load_and_save_is_identity(self):
for backend in BACKENDS:
if backend == 'sox_io':
continue
with self.subTest():
torchaudio.set_audio_backend(backend)
self._test_3_load_and_save_is_identity()
def _test_3_load_and_save_is_identity(self):
input_path = os.path.join(self.test_dirpath, 'assets', 'sinewave.wav')
tensor, sample_rate = torchaudio.load(input_path)
output_path = os.path.join(self.test_dirpath, 'test.wav')
torchaudio.save(output_path, tensor, sample_rate)
tensor2, sample_rate2 = torchaudio.load(output_path)
self.assertTrue(tensor.allclose(tensor2))
self.assertEqual(sample_rate, sample_rate2)
os.unlink(output_path)
@unittest.skipIf(any(be not in BACKENDS for be in ["sox", "soundfile"]), "sox and soundfile are not available")
def test_3_load_and_save_is_identity_across_backend(self):
with self.subTest():
self._test_3_load_and_save_is_identity_across_backend("sox", "soundfile")
with self.subTest():
self._test_3_load_and_save_is_identity_across_backend("soundfile", "sox")
def _test_3_load_and_save_is_identity_across_backend(self, backend1, backend2):
torchaudio.set_audio_backend(backend1)
input_path = os.path.join(self.test_dirpath, 'assets', 'sinewave.wav')
tensor1, sample_rate1 = torchaudio.load(input_path)
output_path = os.path.join(self.test_dirpath, 'test.wav')
torchaudio.save(output_path, tensor1, sample_rate1)
torchaudio.set_audio_backend(backend2)
tensor2, sample_rate2 = torchaudio.load(output_path)
self.assertTrue(tensor1.allclose(tensor2))
self.assertEqual(sample_rate1, sample_rate2)
os.unlink(output_path)
def test_4_load_partial(self):
for backend in BACKENDS_MP3:
with self.subTest():
torchaudio.set_audio_backend(backend)
self._test_4_load_partial()
def _test_4_load_partial(self):
num_frames = 101
offset = 201
# load entire mono sinewave wav file, load a partial copy and then compare
input_sine_path = os.path.join(self.test_dirpath, 'assets', 'sinewave.wav')
x_sine_full, sr_sine = torchaudio.load(input_sine_path)
x_sine_part, _ = torchaudio.load(input_sine_path, num_frames=num_frames, offset=offset)
l1_error = x_sine_full[:, offset:(num_frames + offset)].sub(x_sine_part).abs().sum().item()
# test for the correct number of samples and that the correct portion was loaded
self.assertEqual(x_sine_part.size(1), num_frames)
self.assertEqual(l1_error, 0.)
# create a two channel version of this wavefile
x_2ch_sine = x_sine_full.repeat(1, 2)
out_2ch_sine_path = os.path.join(self.test_dirpath, 'assets', '2ch_sinewave.wav')
torchaudio.save(out_2ch_sine_path, x_2ch_sine, sr_sine)
x_2ch_sine_load, _ = torchaudio.load(out_2ch_sine_path, num_frames=num_frames, offset=offset)
os.unlink(out_2ch_sine_path)
l1_error = x_2ch_sine_load.sub(x_2ch_sine[:, offset:(offset + num_frames)]).abs().sum().item()
self.assertEqual(l1_error, 0.)
# test with two channel mp3
x_2ch_full, sr_2ch = torchaudio.load(self.test_filepath, normalization=True)
x_2ch_part, _ = torchaudio.load(self.test_filepath, normalization=True, num_frames=num_frames, offset=offset)
l1_error = x_2ch_full[:, offset:(offset + num_frames)].sub(x_2ch_part).abs().sum().item()
self.assertEqual(x_2ch_part.size(1), num_frames)
self.assertEqual(l1_error, 0.)
# check behavior if number of samples would exceed file length
offset_ns = 300
x_ns, _ = torchaudio.load(input_sine_path, num_frames=100000, offset=offset_ns)
self.assertEqual(x_ns.size(1), x_sine_full.size(1) - offset_ns)
# check when offset is beyond the end of the file
with self.assertRaises(RuntimeError):
torchaudio.load(input_sine_path, offset=100000)
def test_5_get_info(self):
for backend in BACKENDS:
with self.subTest():
torchaudio.set_audio_backend(backend)
self._test_5_get_info()
def _test_5_get_info(self):
input_path = os.path.join(self.test_dirpath, 'assets', 'sinewave.wav')
channels, samples, rate, precision = (1, 64000, 16000, 16)
si, ei = torchaudio.info(input_path)
self.assertEqual(si.channels, channels)
self.assertEqual(si.length, samples)
self.assertEqual(si.rate, rate)
self.assertEqual(ei.bits_per_sample, precision)
test/torchaudio_unittest/backend/soundfile/info_test.py
View file @ 33dc817c
...@@ -3,7 +3,7 @@ import warnings ...@@ -3,7 +3,7 @@ import warnings
import tarfile import tarfile
import torch import torch
from torchaudio.backend import _soundfile_backend as soundfile_backend from torchaudio.backend import soundfile_backend
from torchaudio._internal import module_utils as _mod_utils from torchaudio._internal import module_utils as _mod_utils
from torchaudio_unittest.common_utils import ( from torchaudio_unittest.common_utils import (
......
test/torchaudio_unittest/backend/soundfile/load_test.py
View file @ 33dc817c
...@@ -4,7 +4,7 @@ from unittest.mock import patch ...@@ -4,7 +4,7 @@ from unittest.mock import patch
import torch import torch
from torchaudio._internal import module_utils as _mod_utils from torchaudio._internal import module_utils as _mod_utils
from torchaudio.backend import _soundfile_backend as soundfile_backend from torchaudio.backend import soundfile_backend
from parameterized import parameterized from parameterized import parameterized
from torchaudio_unittest.common_utils import ( from torchaudio_unittest.common_utils import (
......
test/torchaudio_unittest/backend/soundfile/save_test.py
View file @ 33dc817c
...@@ -2,7 +2,7 @@ import io ...@@ -2,7 +2,7 @@ import io
from unittest.mock import patch from unittest.mock import patch
from torchaudio._internal import module_utils as _mod_utils from torchaudio._internal import module_utils as _mod_utils
from torchaudio.backend import _soundfile_backend as soundfile_backend from torchaudio.backend import soundfile_backend
from torchaudio_unittest.common_utils import ( from torchaudio_unittest.common_utils import (
TempDirMixin, TempDirMixin,
......
test/torchaudio_unittest/backend/utils_test.py
View file @ 33dc817c
...@@ -25,31 +25,13 @@ class TestBackendSwitch_NoBackend(BackendSwitchMixin, common_utils.TorchaudioTes ...@@ -25,31 +25,13 @@ class TestBackendSwitch_NoBackend(BackendSwitchMixin, common_utils.TorchaudioTes
backend_module = torchaudio.backend.no_backend backend_module = torchaudio.backend.no_backend
@common_utils.skipIfNoExtension
class TestBackendSwitch_SoX(BackendSwitchMixin, common_utils.TorchaudioTestCase):
backend = 'sox'
backend_module = torchaudio.backend.sox_backend
@common_utils.skipIfNoExtension @common_utils.skipIfNoExtension
class TestBackendSwitch_SoXIO(BackendSwitchMixin, common_utils.TorchaudioTestCase): class TestBackendSwitch_SoXIO(BackendSwitchMixin, common_utils.TorchaudioTestCase):
backend = 'sox_io' backend = 'sox_io'
backend_module = torchaudio.backend.sox_io_backend backend_module = torchaudio.backend.sox_io_backend
@common_utils.skipIfNoModule('soundfile')
class TestBackendSwitch_soundfile_legacy(BackendSwitchMixin, common_utils.TorchaudioTestCase):
backend = 'soundfile'
backend_module = torchaudio.backend.soundfile_backend
def setUp(self):
torchaudio.USE_SOUNDFILE_LEGACY_INTERFACE = True
def tearDown(self):
torchaudio.USE_SOUNDFILE_LEGACY_INTERFACE = None
@common_utils.skipIfNoModule('soundfile') @common_utils.skipIfNoModule('soundfile')
class TestBackendSwitch_soundfile(BackendSwitchMixin, common_utils.TorchaudioTestCase): class TestBackendSwitch_soundfile(BackendSwitchMixin, common_utils.TorchaudioTestCase):
backend = 'soundfile' backend = 'soundfile'
backend_module = torchaudio.backend._soundfile_backend backend_module = torchaudio.backend.soundfile_backend
test/torchaudio_unittest/common_utils/backend_utils.py
View file @ 33dc817c
...@@ -6,15 +6,13 @@ import torchaudio ...@@ -6,15 +6,13 @@ import torchaudio
def set_audio_backend(backend): def set_audio_backend(backend):
"""Allow additional backend value, 'default'""" """Allow additional backend value, 'default'"""
backends = torchaudio.list_audio_backends() backends = torchaudio.list_audio_backends()
if backend == 'soundfile-new': if backend == 'soundfile':
be = 'soundfile' be = 'soundfile'
torchaudio.USE_SOUNDFILE_LEGACY_INTERFACE = False
elif backend == 'default': elif backend == 'default':
if 'sox_io' in backends: if 'sox_io' in backends:
be = 'sox_io' be = 'sox_io'
elif 'soundfile' in backends: elif 'soundfile' in backends:
be = 'soundfile' be = 'soundfile'
torchaudio.USE_SOUNDFILE_LEGACY_INTERFACE = True
else: else:
raise unittest.SkipTest('No default backend available') raise unittest.SkipTest('No default backend available')
else: else:
......
test/torchaudio_unittest/datasets/tedlium_test.py
View file @ 33dc817c
...@@ -143,13 +143,6 @@ class TestTedliumSoundfile(Tedlium, TorchaudioTestCase): ...@@ -143,13 +143,6 @@ class TestTedliumSoundfile(Tedlium, TorchaudioTestCase):
backend = "soundfile" backend = "soundfile"
class TestTedliumSoundfileNew(Tedlium, TorchaudioTestCase):
backend = "soundfile-new"
if platform.system() != "Windows": if platform.system() != "Windows":
class TestTedliumSox(Tedlium, TorchaudioTestCase):
backend = "sox"
class TestTedliumSoxIO(Tedlium, TorchaudioTestCase): class TestTedliumSoxIO(Tedlium, TorchaudioTestCase):
backend = "sox_io" backend = "sox_io"
torchaudio/__init__.py
View file @ 33dc817c
...@@ -10,20 +10,10 @@ from torchaudio import ( ...@@ -10,20 +10,10 @@ from torchaudio import (
transforms, transforms,
) )
USE_SOUNDFILE_LEGACY_INTERFACE = None
from torchaudio.backend import ( from torchaudio.backend import (
list_audio_backends, list_audio_backends,
get_audio_backend, get_audio_backend,
set_audio_backend, set_audio_backend,
save_encinfo,
sox_signalinfo_t,
sox_encodinginfo_t,
get_sox_option_t,
get_sox_encoding_t,
get_sox_bool,
SignalInfo,
EncodingInfo,
) )
try: try:
......
torchaudio/backend/__init__.py
View file @ 33dc817c
...@@ -5,18 +5,6 @@ from .utils import ( ...@@ -5,18 +5,6 @@ from .utils import (
get_audio_backend, get_audio_backend,
set_audio_backend, set_audio_backend,
) )
from .sox_backend import (
save_encinfo,
sox_signalinfo_t,
sox_encodinginfo_t,
get_sox_option_t,
get_sox_encoding_t,
get_sox_bool,
)
from .common import (
SignalInfo,
EncodingInfo,
)
utils._init_audio_backend() utils._init_audio_backend()
torchaudio/backend/_soundfile_backend.py deleted 100644 → 0
View file @ 086467a1
"""The new soundfile backend which will become default in 0.8.0 onward"""
from typing import Tuple, Optional
import warnings
import torch
from torchaudio._internal import module_utils as _mod_utils
from .common import AudioMetaData
if _mod_utils.is_module_available("soundfile"):
import soundfile
# Mapping from soundfile subtype to number of bits per sample.
# This is mostly heuristical and the value is set to 0 when it is irrelevant
# (lossy formats) or when it can't be inferred.
# For ADPCM (and G72X) subtypes, it's hard to infer the bit depth because it's not part of the standard:
# According to https://en.wikipedia.org/wiki/Adaptive_differential_pulse-code_modulation#In_telephony,
# the default seems to be 8 bits but it can be compressed further to 4 bits.
# The dict is inspired from
# https://github.com/bastibe/python-soundfile/blob/744efb4b01abc72498a96b09115b42a4cabd85e4/soundfile.py#L66-L94
_SUBTYPE_TO_BITS_PER_SAMPLE = {
'PCM_S8': 8, # Signed 8 bit data
'PCM_16': 16, # Signed 16 bit data
'PCM_24': 24, # Signed 24 bit data
'PCM_32': 32, # Signed 32 bit data
'PCM_U8': 8, # Unsigned 8 bit data (WAV and RAW only)
'FLOAT': 32, # 32 bit float data
'DOUBLE': 64, # 64 bit float data
'ULAW': 8, # U-Law encoded. See https://en.wikipedia.org/wiki/G.711#Types
'ALAW': 8, # A-Law encoded. See https://en.wikipedia.org/wiki/G.711#Types
'IMA_ADPCM': 0, # IMA ADPCM.
'MS_ADPCM': 0, # Microsoft ADPCM.
'GSM610': 0, # GSM 6.10 encoding. (Wikipedia says 1.625 bit depth?? https://en.wikipedia.org/wiki/Full_Rate)
'VOX_ADPCM': 0, # OKI / Dialogix ADPCM
'G721_32': 0, # 32kbs G721 ADPCM encoding.
'G723_24': 0, # 24kbs G723 ADPCM encoding.
'G723_40': 0, # 40kbs G723 ADPCM encoding.
'DWVW_12': 12, # 12 bit Delta Width Variable Word encoding.
'DWVW_16': 16, # 16 bit Delta Width Variable Word encoding.
'DWVW_24': 24, # 24 bit Delta Width Variable Word encoding.
'DWVW_N': 0, # N bit Delta Width Variable Word encoding.
'DPCM_8': 8, # 8 bit differential PCM (XI only)
'DPCM_16': 16, # 16 bit differential PCM (XI only)
'VORBIS': 0, # Xiph Vorbis encoding. (lossy)
'ALAC_16': 16, # Apple Lossless Audio Codec (16 bit).
'ALAC_20': 20, # Apple Lossless Audio Codec (20 bit).
'ALAC_24': 24, # Apple Lossless Audio Codec (24 bit).
'ALAC_32': 32, # Apple Lossless Audio Codec (32 bit).
}
def _get_bit_depth(subtype):
if subtype not in _SUBTYPE_TO_BITS_PER_SAMPLE:
warnings.warn(
f"The {subtype} subtype is unknown to TorchAudio. As a result, the bits_per_sample "
"attribute will be set to 0. If you are seeing this warning, please "
"report by opening an issue on github (after checking for existing/closed ones). "
"You may otherwise ignore this warning."
)
return _SUBTYPE_TO_BITS_PER_SAMPLE.get(subtype, 0)
_SUBTYPE_TO_ENCODING = {
'PCM_S8': 'PCM_S',
'PCM_16': 'PCM_S',
'PCM_24': 'PCM_S',
'PCM_32': 'PCM_S',
'PCM_U8': 'PCM_U',
'FLOAT': 'PCM_F',
'DOUBLE': 'PCM_F',
'ULAW': 'ULAW',
'ALAW': 'ALAW',
'VORBIS': 'VORBIS',
}
def _get_encoding(format: str, subtype: str):
if format == 'FLAC':
return 'FLAC'
return _SUBTYPE_TO_ENCODING.get(subtype, 'UNKNOWN')
@_mod_utils.requires_module("soundfile")
def info(filepath: str, format: Optional[str] = None) -> AudioMetaData:
"""Get signal information of an audio file.
Args:
filepath (path-like object or file-like object):
Source of audio data.
Note:
* This argument is intentionally annotated as ``str`` only,
for the consistency with "sox_io" backend, which has a restriction
on type annotation due to TorchScript compiler compatiblity.
format (str, optional):
Not used. PySoundFile does not accept format hint.
Returns:
AudioMetaData: meta data of the given audio.
"""
sinfo = soundfile.info(filepath)
return AudioMetaData(
sinfo.samplerate,
sinfo.frames,
sinfo.channels,
bits_per_sample=_get_bit_depth(sinfo.subtype),
encoding=_get_encoding(sinfo.format, sinfo.subtype),
)
_SUBTYPE2DTYPE = {
"PCM_S8": "int8",
"PCM_U8": "uint8",
"PCM_16": "int16",
"PCM_32": "int32",
"FLOAT": "float32",
"DOUBLE": "float64",
}
@_mod_utils.requires_module("soundfile")
def load(
filepath: str,
frame_offset: int = 0,
num_frames: int = -1,
normalize: bool = True,
channels_first: bool = True,
format: Optional[str] = None,
) -> Tuple[torch.Tensor, int]:
"""Load audio data from file.
Note:
The formats this function can handle depend on the soundfile installation.
This function is tested on the following formats;
* WAV
* 32-bit floating-point
* 32-bit signed integer
* 16-bit signed integer
* 8-bit unsigned integer
* FLAC
* OGG/VORBIS
* SPHERE
By default (``normalize=True``, ``channels_first=True``), this function returns Tensor with
``float32`` dtype and the shape of ``[channel, time]``.
The samples are normalized to fit in the range of ``[-1.0, 1.0]``.
When the input format is WAV with integer type, such as 32-bit signed integer, 16-bit
signed integer and 8-bit unsigned integer (24-bit signed integer is not supported),
by providing ``normalize=False``, this function can return integer Tensor, where the samples
are expressed within the whole range of the corresponding dtype, that is, ``int32`` tensor
for 32-bit signed PCM, ``int16`` for 16-bit signed PCM and ``uint8`` for 8-bit unsigned PCM.
``normalize`` parameter has no effect on 32-bit floating-point WAV and other formats, such as
``flac`` and ``mp3``.
For these formats, this function always returns ``float32`` Tensor with values normalized to
``[-1.0, 1.0]``.
Args:
filepath (path-like object or file-like object):
Source of audio data.
Note:
* This argument is intentionally annotated as ``str`` only,
for the consistency with "sox_io" backend, which has a restriction
on type annotation due to TorchScript compiler compatiblity.
frame_offset (int):
Number of frames to skip before start reading data.
num_frames (int):
Maximum number of frames to read. ``-1`` reads all the remaining samples,
starting from ``frame_offset``.
This function may return the less number of frames if there is not enough
frames in the given file.
normalize (bool):
When ``True``, this function always return ``float32``, and sample values are
normalized to ``[-1.0, 1.0]``.
If input file is integer WAV, giving ``False`` will change the resulting Tensor type to
integer type.
This argument has no effect for formats other than integer WAV type.
channels_first (bool):
When True, the returned Tensor has dimension ``[channel, time]``.
Otherwise, the returned Tensor's dimension is ``[time, channel]``.
format (str, optional):
Not used. PySoundFile does not accept format hint.
Returns:
Tuple[torch.Tensor, int]: Resulting Tensor and sample rate.
If the input file has integer wav format and normalization is off, then it has
integer type, else ``float32`` type. If ``channels_first=True``, it has
``[channel, time]`` else ``[time, channel]``.
"""
with soundfile.SoundFile(filepath, "r") as file_:
if file_.format != "WAV" or normalize:
dtype = "float32"
elif file_.subtype not in _SUBTYPE2DTYPE:
raise ValueError(f"Unsupported subtype: {file_.subtype}")
else:
dtype = _SUBTYPE2DTYPE[file_.subtype]
frames = file_._prepare_read(frame_offset, None, num_frames)
waveform = file_.read(frames, dtype, always_2d=True)
sample_rate = file_.samplerate
waveform = torch.from_numpy(waveform)
if channels_first:
waveform = waveform.t()
return waveform, sample_rate
def _get_subtype_for_wav(
dtype: torch.dtype,
encoding: str,
bits_per_sample: int):
if not encoding:
if not bits_per_sample:
subtype = {
torch.uint8: "PCM_U8",
torch.int16: "PCM_16",
torch.int32: "PCM_32",
torch.float32: "FLOAT",
torch.float64: "DOUBLE",
}.get(dtype)
if not subtype:
raise ValueError(f"Unsupported dtype for wav: {dtype}")
return subtype
if bits_per_sample == 8:
return "PCM_U8"
return f"PCM_{bits_per_sample}"
if encoding == "PCM_S":
if not bits_per_sample:
return "PCM_32"
if bits_per_sample == 8:
raise ValueError("wav does not support 8-bit signed PCM encoding.")
return f"PCM_{bits_per_sample}"
if encoding == "PCM_U":
if bits_per_sample in (None, 8):
return "PCM_U8"
raise ValueError("wav only supports 8-bit unsigned PCM encoding.")
if encoding == "PCM_F":
if bits_per_sample in (None, 32):
return "FLOAT"
if bits_per_sample == 64:
return "DOUBLE"
raise ValueError("wav only supports 32/64-bit float PCM encoding.")
if encoding == "ULAW":
if bits_per_sample in (None, 8):
return "ULAW"
raise ValueError("wav only supports 8-bit mu-law encoding.")
if encoding == "ALAW":
if bits_per_sample in (None, 8):
return "ALAW"
raise ValueError("wav only supports 8-bit a-law encoding.")
raise ValueError(f"wav does not support {encoding}.")
def _get_subtype_for_sphere(encoding: str, bits_per_sample: int):
if encoding in (None, "PCM_S"):
return f"PCM_{bits_per_sample}" if bits_per_sample else "PCM_32"
if encoding in ("PCM_U", "PCM_F"):
raise ValueError(f"sph does not support {encoding} encoding.")
if encoding == "ULAW":
if bits_per_sample in (None, 8):
return "ULAW"
raise ValueError("sph only supports 8-bit for mu-law encoding.")
if encoding == "ALAW":
return "ALAW"
raise ValueError(f"sph does not support {encoding}.")
def _get_subtype(
dtype: torch.dtype,
format: str,
encoding: str,
bits_per_sample: int):
if format == "wav":
return _get_subtype_for_wav(dtype, encoding, bits_per_sample)
if format == "flac":
if encoding:
raise ValueError("flac does not support encoding.")
if not bits_per_sample:
return "PCM_24"
if bits_per_sample > 24:
raise ValueError("flac does not support bits_per_sample > 24.")
return "PCM_S8" if bits_per_sample == 8 else f"PCM_{bits_per_sample}"
if format in ("ogg", "vorbis"):
if encoding or bits_per_sample:
raise ValueError(
"ogg/vorbis does not support encoding/bits_per_sample.")
return "VORBIS"
if format == "sph":
return _get_subtype_for_sphere(encoding, bits_per_sample)
if format in ("nis", "nist"):
return "PCM_16"
raise ValueError(f"Unsupported format: {format}")
@_mod_utils.requires_module("soundfile")
def save(
filepath: str,
src: torch.Tensor,
sample_rate: int,
channels_first: bool = True,
compression: Optional[float] = None,
format: Optional[str] = None,
encoding: Optional[str] = None,
bits_per_sample: Optional[int] = None,
):
"""Save audio data to file.
Note:
The formats this function can handle depend on the soundfile installation.
This function is tested on the following formats;
* WAV
* 32-bit floating-point
* 32-bit signed integer
* 16-bit signed integer
* 8-bit unsigned integer
* FLAC
* OGG/VORBIS
* SPHERE
Args:
filepath (str or pathlib.Path): Path to audio file.
This functionalso handles ``pathlib.Path`` objects, but is annotated as ``str``
for the consistency with "sox_io" backend, which has a restriction on type annotation
for TorchScript compiler compatiblity.
src (torch.Tensor): Audio data to save. must be 2D tensor.
sample_rate (int): sampling rate
channels_first (bool): If ``True``, the given tensor is interpreted as ``[channel, time]``,
otherwise ``[time, channel]``.
compression (Optional[float]): Not used.
It is here only for interface compatibility reson with "sox_io" backend.
format (str, optional): Override the audio format.
When ``filepath`` argument is path-like object, audio format is
inferred from file extension. If the file extension is missing or
different, you can specify the correct format with this argument.
When ``filepath`` argument is file-like object,
this argument is required.
Valid values are ``"wav"``, ``"ogg"``, ``"vorbis"``,
``"flac"`` and ``"sph"``.
encoding (str, optional): Changes the encoding for supported formats.
This argument is effective only for supported formats, sush as
``"wav"``, ``""flac"`` and ``"sph"``. Valid values are;
- ``"PCM_S"`` (signed integer Linear PCM)
- ``"PCM_U"`` (unsigned integer Linear PCM)
- ``"PCM_F"`` (floating point PCM)
- ``"ULAW"`` (mu-law)
- ``"ALAW"`` (a-law)
bits_per_sample (int, optional): Changes the bit depth for the
supported formats.
When ``format`` is one of ``"wav"``, ``"flac"`` or ``"sph"``,
you can change the bit depth.
Valid values are ``8``, ``16``, ``24``, ``32`` and ``64``.
Supported formats/encodings/bit depth/compression are:
``"wav"``
- 32-bit floating-point PCM
- 32-bit signed integer PCM
- 24-bit signed integer PCM
- 16-bit signed integer PCM
- 8-bit unsigned integer PCM
- 8-bit mu-law
- 8-bit a-law
Note: Default encoding/bit depth is determined by the dtype of
the input Tensor.
``"flac"``
- 8-bit
- 16-bit
- 24-bit (default)
``"ogg"``, ``"vorbis"``
- Doesn't accept changing configuration.
``"sph"``
- 8-bit signed integer PCM
- 16-bit signed integer PCM
- 24-bit signed integer PCM
- 32-bit signed integer PCM (default)
- 8-bit mu-law
- 8-bit a-law
- 16-bit a-law
- 24-bit a-law
- 32-bit a-law
"""
if src.ndim != 2:
raise ValueError(f"Expected 2D Tensor, got {src.ndim}D.")
if compression is not None:
warnings.warn(
'`save` function of "soundfile" backend does not support "compression" parameter. '
"The argument is silently ignored."
)
if hasattr(filepath, 'write'):
if format is None:
raise RuntimeError('`format` is required when saving to file object.')
ext = format.lower()
else:
ext = str(filepath).split(".")[-1].lower()
if bits_per_sample not in (None, 8, 16, 24, 32, 64):
raise ValueError("Invalid bits_per_sample.")
subtype = _get_subtype(src.dtype, ext, encoding, bits_per_sample)
# sph is a extension used in TED-LIUM but soundfile does not recognize it as NIST format,
# so we extend the extensions manually here
if ext in ["nis", "nist", "sph"] and format is None:
format = "NIST"
if channels_first:
src = src.t()
soundfile.write(
file=filepath, data=src, samplerate=sample_rate, subtype=subtype, format=format
)
@_mod_utils.requires_module("soundfile")
@_mod_utils.deprecated('Please use "torchaudio.load".', "0.9.0")
def load_wav(
filepath: str,
frame_offset: int = 0,
num_frames: int = -1,
channels_first: bool = True,
) -> Tuple[torch.Tensor, int]:
"""Load wave file.
This function is defined only for the purpose of compatibility against other backend
for simple usecases, such as ``torchaudio.load_wav(filepath)``.
The implementation is same as :py:func:`load`.
"""
return load(
filepath,
frame_offset,
num_frames,
normalize=False,
channels_first=channels_first,
)
torchaudio/backend/common.py
View file @ 33dc817c
from typing import Any, Optional
import warnings
class AudioMetaData: class AudioMetaData:
"""Return type of ``torchaudio.info`` function. """Return type of ``torchaudio.info`` function.
...@@ -28,185 +24,3 @@ class AudioMetaData: ...@@ -28,185 +24,3 @@ class AudioMetaData:
self.num_channels = num_channels self.num_channels = num_channels
self.bits_per_sample = bits_per_sample self.bits_per_sample = bits_per_sample
self.encoding = encoding self.encoding = encoding
class SignalInfo:
"""One of return types of ``torchaudio.info`` functions.
This class is used by :ref:`"sox" backend (deprecated)<sox_backend>` and
:ref:`"soundfile" backend with the legacy interface (deprecated)<soundfile_legacy_backend>`.
See https://fossies.org/dox/sox-14.4.2/structsox__signalinfo__t.html
:ivar Optional[int] channels: The number of channels
:ivar Optional[float] rate: Sampleing rate
:ivar Optional[int] precision: Bit depth
:ivar Optional[int] length: For :ref:`sox backend<sox_backend>`, the number of samples.
(frames * channels). For :ref:`soundfile backend<soundfile_backend>`, the number of frames.
"""
def __init__(self,
channels: Optional[int] = None,
rate: Optional[float] = None,
precision: Optional[int] = None,
length: Optional[int] = None) -> None:
message = (
f'{self.__module__}.{self.__class__.__name__} has been deprecated '
'and will be removed from 0.9.0 release. '
'Please migrate to `AudioMetaData`.'
)
warnings.warn(message)
self.channels = channels
self.rate = rate
self.precision = precision
self.length = length
class EncodingInfo:
"""One of return types of ``torchaudio.info`` functions.
This class is used by :ref:`"sox" backend (deprecated)<sox_backend>` and
:ref:`"soundfile" backend with the legacy interface (deprecated)<soundfile_legacy_backend>`.
See https://fossies.org/dox/sox-14.4.2/structsox__encodinginfo__t.html
:ivar Optional[int] encoding: sox_encoding_t
:ivar Optional[int] bits_per_sample: bit depth
:ivar Optional[float] compression: Compression option
:ivar Any reverse_bytes:
:ivar Any reverse_nibbles:
:ivar Any reverse_bits:
:ivar Optional[bool] opposite_endian:
"""
def __init__(self,
encoding: Any = None,
bits_per_sample: Optional[int] = None,
compression: Optional[float] = None,
reverse_bytes: Any = None,
reverse_nibbles: Any = None,
reverse_bits: Any = None,
opposite_endian: Optional[bool] = None) -> None:
message = (
f'{self.__module__}.{self.__class__.__name__} has been deprecated '
'and will be removed from 0.9.0 release. '
'Please migrate to `AudioMetaData`.'
)
warnings.warn(message)
self.encoding = encoding
self.bits_per_sample = bits_per_sample
self.compression = compression
self.reverse_bytes = reverse_bytes
self.reverse_nibbles = reverse_nibbles
self.reverse_bits = reverse_bits
self.opposite_endian = opposite_endian
_LOAD_DOCSTRING = r"""Loads an audio file from disk into a tensor
Args:
filepath: Path to audio file
out: An optional output tensor to use instead of creating one. (Default: ``None``)
normalization: Optional normalization.
If boolean `True`, then output is divided by `1 << 31`.
Assuming the input is signed 32-bit audio, this normalizes to `[-1, 1]`.
If `float`, then output is divided by that number.
If `Callable`, then the output is passed as a paramete to the given function,
then the output is divided by the result. (Default: ``True``)
channels_first: Set channels first or length first in result. (Default: ``True``)
num_frames: Number of frames to load. 0 to load everything after the offset.
(Default: ``0``)
offset: Number of frames from the start of the file to begin data loading.
(Default: ``0``)
signalinfo: A sox_signalinfo_t type, which could be helpful if the
audio type cannot be automatically determined. (Default: ``None``)
encodinginfo: A sox_encodinginfo_t type, which could be set if the
audio type cannot be automatically determined. (Default: ``None``)
filetype: A filetype or extension to be set if sox cannot determine it
automatically. (Default: ``None``)
Returns:
(Tensor, int): An output tensor of size `[C x L]` or `[L x C]` where
L is the number of audio frames and
C is the number of channels.
An integer which is the sample rate of the audio (as listed in the metadata of the file)
Example
>>> data, sample_rate = torchaudio.load('foo.mp3')
>>> print(data.size())
torch.Size([2, 278756])
>>> print(sample_rate)
44100
>>> data_vol_normalized, _ = torchaudio.load('foo.mp3', normalization=lambda x: torch.abs(x).max())
>>> print(data_vol_normalized.abs().max())
1.
"""
_LOAD_WAV_DOCSTRING = r""" Loads a wave file.
It assumes that the wav file uses 16 bit per sample that needs normalization by
shifting the input right by 16 bits.
Args:
filepath: Path to audio file
Returns:
(Tensor, int): An output tensor of size `[C x L]` or `[L x C]` where L is the number
of audio frames and C is the number of channels. An integer which is the sample rate of the
audio (as listed in the metadata of the file)
"""
_SAVE_DOCSTRING = r"""Saves a Tensor on file as an audio file
Args:
filepath: Path to audio file
src: An input 2D tensor of shape `[C x L]` or `[L x C]` where L is
the number of audio frames, C is the number of channels
sample_rate: An integer which is the sample rate of the
audio (as listed in the metadata of the file)
precision Bit precision (Default: ``16``)
channels_first (bool, optional): Set channels first or length first in result. (
Default: ``True``)
"""
_INFO_DOCSTRING = r"""Gets metadata from an audio file without loading the signal.
Args:
filepath: Path to audio file
Returns:
(sox_signalinfo_t, sox_encodinginfo_t): A si (sox_signalinfo_t) signal
info as a python object. An ei (sox_encodinginfo_t) encoding info
Example
>>> si, ei = torchaudio.info('foo.wav')
>>> rate, channels, encoding = si.rate, si.channels, ei.encoding
"""
def _impl_load(func):
setattr(func, '__doc__', _LOAD_DOCSTRING)
return func
def _impl_load_wav(func):
setattr(func, '__doc__', _LOAD_WAV_DOCSTRING)
return func
def _impl_save(func):
setattr(func, '__doc__', _SAVE_DOCSTRING)
return func
def _impl_info(func):
setattr(func, '__doc__', _INFO_DOCSTRING)
return func
torchaudio/backend/no_backend.py
View file @ 33dc817c
...@@ -3,33 +3,24 @@ from typing import Any, Callable, Optional, Tuple, Union ...@@ -3,33 +3,24 @@ from typing import Any, Callable, Optional, Tuple, Union
from torch import Tensor from torch import Tensor
from . import common
from .common import SignalInfo, EncodingInfo
@common._impl_load
def load(filepath: Union[str, Path], def load(filepath: Union[str, Path],
out: Optional[Tensor] = None, out: Optional[Tensor] = None,
normalization: Union[bool, float, Callable] = True, normalization: Union[bool, float, Callable] = True,
channels_first: bool = True, channels_first: bool = True,
num_frames: int = 0, num_frames: int = 0,
offset: int = 0, offset: int = 0,
signalinfo: Optional[SignalInfo] = None,
encodinginfo: Optional[EncodingInfo] = None,
filetype: Optional[str] = None) -> Tuple[Tensor, int]: filetype: Optional[str] = None) -> Tuple[Tensor, int]:
raise RuntimeError('No audio I/O backend is available.') raise RuntimeError('No audio I/O backend is available.')
@common._impl_load_wav
def load_wav(filepath: Union[str, Path], **kwargs: Any) -> Tuple[Tensor, int]: def load_wav(filepath: Union[str, Path], **kwargs: Any) -> Tuple[Tensor, int]:
raise RuntimeError('No audio I/O backend is available.') raise RuntimeError('No audio I/O backend is available.')
@common._impl_save
def save(filepath: str, src: Tensor, sample_rate: int, precision: int = 16, channels_first: bool = True) -> None: def save(filepath: str, src: Tensor, sample_rate: int, precision: int = 16, channels_first: bool = True) -> None:
raise RuntimeError('No audio I/O backend is available.') raise RuntimeError('No audio I/O backend is available.')
@common._impl_info def info(filepath: str) -> None:
def info(filepath: str) -> Tuple[SignalInfo, EncodingInfo]:
raise RuntimeError('No audio I/O backend is available.') raise RuntimeError('No audio I/O backend is available.')
torchaudio/backend/soundfile_backend.py
View file @ 33dc817c
This diff is collapsed.
torchaudio/backend/sox_backend.py deleted 100644 → 0
View file @ 086467a1
import os.path
from typing import Any, Optional, Tuple
import torch
from torch import Tensor
from torchaudio._internal import (
module_utils as _mod_utils,
misc_ops as _misc_ops,
)
from . import common
from .common import SignalInfo, EncodingInfo
if _mod_utils.is_module_available('torchaudio._torchaudio'):
from torchaudio import _torchaudio
@_mod_utils.requires_module('torchaudio._torchaudio')
@common._impl_load
def load(filepath: str,
out: Optional[Tensor] = None,
normalization: bool = True,
channels_first: bool = True,
num_frames: int = 0,
offset: int = 0,
signalinfo: SignalInfo = None,
encodinginfo: EncodingInfo = None,
filetype: Optional[str] = None) -> Tuple[Tensor, int]:
r"""See torchaudio.load"""
# stringify if `pathlib.Path` (noop if already `str`)
filepath = str(filepath)
# check if valid file
if not os.path.isfile(filepath):
raise OSError("{} not found or is a directory".format(filepath))
# initialize output tensor
if out is not None:
_misc_ops.check_input(out)
else:
out = torch.FloatTensor()
if num_frames < -1:
raise ValueError("Expected value for num_samples -1 (entire file) or >=0")
if offset < 0:
raise ValueError("Expected positive offset value")
sample_rate = _torchaudio.read_audio_file(
filepath,
out,
channels_first,
num_frames,
offset,
signalinfo,
encodinginfo,
filetype
)
# normalize if needed
_misc_ops.normalize_audio(out, normalization)
return out, sample_rate
@_mod_utils.requires_module('torchaudio._torchaudio')
@_mod_utils.deprecated('Please use "torchaudio.load".', '0.9.0')
@common._impl_load_wav
def load_wav(filepath, **kwargs):
kwargs['normalization'] = 1 << 16
return load(filepath, **kwargs)
@_mod_utils.requires_module('torchaudio._torchaudio')
@common._impl_save
def save(filepath: str, src: Tensor, sample_rate: int, precision: int = 16, channels_first: bool = True) -> None:
r"""See torchaudio.save"""
si = sox_signalinfo_t()
ch_idx = 0 if channels_first else 1
si.rate = sample_rate
si.channels = 1 if src.dim() == 1 else src.size(ch_idx)
si.length = src.numel()
si.precision = precision
return save_encinfo(filepath, src, channels_first, si)
@_mod_utils.requires_module('torchaudio._torchaudio')
@common._impl_info
def info(filepath: str) -> Tuple[SignalInfo, EncodingInfo]:
r"""See torchaudio.info"""
return _torchaudio.get_info(filepath)
@_mod_utils.requires_module('torchaudio._torchaudio')
@_mod_utils.deprecated(
'Please migrate to "sox_io" backend. See https://github.com/pytorch/audio/issues/903 for the detail',
'0.9.0')
def save_encinfo(filepath: str,
src: Tensor,
channels_first: bool = True,
signalinfo: Optional[SignalInfo] = None,
encodinginfo: Optional[EncodingInfo] = None,
filetype: Optional[str] = None) -> None:
r"""Saves a tensor of an audio signal to disk as a standard format like mp3, wav, etc.
Args:
filepath (str): Path to audio file
src (Tensor): An input 2D tensor of shape `[C x L]` or `[L x C]` where L is
the number of audio frames, C is the number of channels
channels_first (bool, optional): Set channels first or length first in result. (Default: ``True``)
signalinfo (sox_signalinfo_t, optional): A sox_signalinfo_t type, which could be helpful if the
audio type cannot be automatically determined (Default: ``None``).
encodinginfo (sox_encodinginfo_t, optional): A sox_encodinginfo_t type, which could be set if the
audio type cannot be automatically determined (Default: ``None``).
filetype (str, optional): A filetype or extension to be set if sox cannot determine it
automatically. (Default: ``None``)
Example
>>> data, sample_rate = torchaudio.load('foo.mp3')
>>> torchaudio.save('foo.wav', data, sample_rate)
"""
ch_idx, len_idx = (0, 1) if channels_first else (1, 0)
# check if save directory exists
abs_dirpath = os.path.dirname(os.path.abspath(filepath))
if not os.path.isdir(abs_dirpath):
raise OSError("Directory does not exist: {}".format(abs_dirpath))
# check that src is a CPU tensor
_misc_ops.check_input(src)
# Check/Fix shape of source data
if src.dim() == 1:
# 1d tensors as assumed to be mono signals
src.unsqueeze_(ch_idx)
elif src.dim() > 2 or src.size(ch_idx) > 16:
# assumes num_channels < 16
raise ValueError(
"Expected format where C < 16, but found {}".format(src.size()))
# sox stores the sample rate as a float, though practically sample rates are almost always integers
# convert integers to floats
if signalinfo:
if signalinfo.rate and not isinstance(signalinfo.rate, float):
if float(signalinfo.rate) == signalinfo.rate:
signalinfo.rate = float(signalinfo.rate)
else:
raise TypeError('Sample rate should be a float or int')
# check if the bit precision (i.e. bits per sample) is an integer
if signalinfo.precision and not isinstance(signalinfo.precision, int):
if int(signalinfo.precision) == signalinfo.precision:
signalinfo.precision = int(signalinfo.precision)
else:
raise TypeError('Bit precision should be an integer')
# programs such as librosa normalize the signal, unnormalize if detected
if src.min() >= -1.0 and src.max() <= 1.0:
src = src * (1 << 31)
src = src.long()
# set filetype and allow for files with no extensions
extension = os.path.splitext(filepath)[1]
filetype = extension[1:] if len(extension) > 0 else filetype
# transpose from C x L -> L x C
if channels_first:
src = src.transpose(1, 0)
# save data to file
src = src.contiguous()
_torchaudio.write_audio_file(filepath, src, signalinfo, encodinginfo, filetype)
@_mod_utils.requires_module('torchaudio._torchaudio')
@_mod_utils.deprecated(
'Please migrate to "sox_io" backend. See https://github.com/pytorch/audio/issues/903 for the detail',
'0.9.0')
def sox_signalinfo_t() -> SignalInfo:
r"""Create a sox_signalinfo_t object. This object can be used to set the sample
rate, number of channels, length, bit precision and headroom multiplier
primarily for effects
Returns: sox_signalinfo_t(object)
- rate (float), sample rate as a float, practically will likely be an integer float
- channel (int), number of audio channels
- precision (int), bit precision
- length (int), length of audio in samples * channels, 0 for unspecified and -1 for unknown
- mult (float, optional), headroom multiplier for effects and ``None`` for no multiplier
Example
>>> si = torchaudio.sox_signalinfo_t()
>>> si.channels = 1
>>> si.rate = 16000.
>>> si.precision = 16
>>> si.length = 0
"""
return _torchaudio.sox_signalinfo_t()
@_mod_utils.requires_module('torchaudio._torchaudio')
@_mod_utils.deprecated(
'Please migrate to "sox_io" backend. See https://github.com/pytorch/audio/issues/903 for the detail',
'0.9.0')
def sox_encodinginfo_t() -> EncodingInfo:
r"""Create a sox_encodinginfo_t object. This object can be used to set the encoding
type, bit precision, compression factor, reverse bytes, reverse nibbles,
reverse bits and endianness. This can be used in an effects chain to encode the
final output or to save a file with a specific encoding. For example, one could
use the sox ulaw encoding to do 8-bit ulaw encoding. Note in a tensor output
the result will be a 32-bit number, but number of unique values will be determined by
the bit precision.
Returns: sox_encodinginfo_t(object)
- encoding (sox_encoding_t), output encoding
- bits_per_sample (int), bit precision, same as `precision` in sox_signalinfo_t
- compression (float), compression for lossy formats, 0.0 for default compression
- reverse_bytes (sox_option_t), reverse bytes, use sox_option_default
- reverse_nibbles (sox_option_t), reverse nibbles, use sox_option_default
- reverse_bits (sox_option_t), reverse bytes, use sox_option_default
- opposite_endian (sox_bool), change endianness, use sox_false
Example
>>> ei = torchaudio.sox_encodinginfo_t()
>>> ei.encoding = torchaudio.get_sox_encoding_t(1)
>>> ei.bits_per_sample = 16
>>> ei.compression = 0
>>> ei.reverse_bytes = torchaudio.get_sox_option_t(2)
>>> ei.reverse_nibbles = torchaudio.get_sox_option_t(2)
>>> ei.reverse_bits = torchaudio.get_sox_option_t(2)
>>> ei.opposite_endian = torchaudio.get_sox_bool(0)
"""
ei = _torchaudio.sox_encodinginfo_t()
sdo = get_sox_option_t(2) # sox_default_option
ei.reverse_bytes = sdo
ei.reverse_nibbles = sdo
ei.reverse_bits = sdo
return ei
@_mod_utils.requires_module('torchaudio._torchaudio')
@_mod_utils.deprecated(
'Please migrate to "sox_io" backend. See https://github.com/pytorch/audio/issues/903 for the detail',
'0.9.0')
def get_sox_encoding_t(i: int = None) -> EncodingInfo:
r"""Get enum of sox_encoding_t for sox encodings.
Args:
i (int, optional): Choose type or get a dict with all possible options
use ``__members__`` to see all options when not specified. (Default: ``None``)
Returns:
sox_encoding_t: A sox_encoding_t type for output encoding
"""
if i is None:
# one can see all possible values using the .__members__ attribute
return _torchaudio.sox_encoding_t
else:
return _torchaudio.sox_encoding_t(i)
@_mod_utils.requires_module('torchaudio._torchaudio')
@_mod_utils.deprecated(
'Please migrate to "sox_io" backend. See https://github.com/pytorch/audio/issues/903 for the detail',
'0.9.0')
def get_sox_option_t(i: int = 2) -> Any:
r"""Get enum of sox_option_t for sox encodinginfo options.
Args:
i (int, optional): Choose type or get a dict with all possible options
use ``__members__`` to see all options when not specified.
(Default: ``sox_option_default`` or ``2``)
Returns:
sox_option_t: A sox_option_t type
"""
if i is None:
return _torchaudio.sox_option_t
else:
return _torchaudio.sox_option_t(i)
@_mod_utils.requires_module('torchaudio._torchaudio')
@_mod_utils.deprecated(
'Please migrate to "sox_io" backend. See https://github.com/pytorch/audio/issues/903 for the detail',
'0.9.0')
def get_sox_bool(i: int = 0) -> Any:
r"""Get enum of sox_bool for sox encodinginfo options.
Args:
i (int, optional): Choose type or get a dict with all possible options
use ``__members__`` to see all options when not specified. (Default:
``sox_false`` or ``0``)
Returns:
sox_bool: A sox_bool type
"""
if i is None:
return _torchaudio.sox_bool
else:
return _torchaudio.sox_bool(i)
torchaudio/backend/utils.py
View file @ 33dc817c
...@@ -6,10 +6,8 @@ import torchaudio ...@@ -6,10 +6,8 @@ import torchaudio
from torchaudio._internal.module_utils import is_module_available from torchaudio._internal.module_utils import is_module_available
from . import ( from . import (
no_backend, no_backend,
sox_backend,
sox_io_backend, sox_io_backend,
soundfile_backend, soundfile_backend,
_soundfile_backend,
) )
__all__ = [ __all__ = [
...@@ -29,7 +27,6 @@ def list_audio_backends() -> List[str]: ...@@ -29,7 +27,6 @@ def list_audio_backends() -> List[str]:
if is_module_available('soundfile'): if is_module_available('soundfile'):
backends.append('soundfile') backends.append('soundfile')
if is_module_available('torchaudio._torchaudio'): if is_module_available('torchaudio._torchaudio'):
backends.append('sox')
backends.append('sox_io') backends.append('sox_io')
return backends return backends
...@@ -39,15 +36,9 @@ def set_audio_backend(backend: Optional[str]): ...@@ -39,15 +36,9 @@ def set_audio_backend(backend: Optional[str]):
Args: Args:
backend (Optional[str]): Name of the backend. backend (Optional[str]): Name of the backend.
One of ``"sox"``, ``"sox_io"`` or ``"soundfile"`` based on availability One of ``"sox_io"`` or ``"soundfile"`` based on availability
of the system. If ``None`` is provided the current backend is unassigned. of the system. If ``None`` is provided the current backend is unassigned.
""" """
if torchaudio.USE_SOUNDFILE_LEGACY_INTERFACE is not None:
warnings.warn(
'"torchaudio.USE_SOUNDFILE_LEGACY_INTERFACE" flag is deprecated and will be removed in 0.9.0. '
'Please remove the use of flag.'
)
if backend is not None and backend not in list_audio_backends(): if backend is not None and backend not in list_audio_backends():
raise RuntimeError( raise RuntimeError(
f'Backend "{backend}" is not one of ' f'Backend "{backend}" is not one of '
...@@ -55,23 +46,10 @@ def set_audio_backend(backend: Optional[str]): ...@@ -55,23 +46,10 @@ def set_audio_backend(backend: Optional[str]):
if backend is None: if backend is None:
module = no_backend module = no_backend
elif backend == 'sox':
warnings.warn(
'"sox" backend is deprecated and will be removed in 0.9.0. '
'Please use "sox_io" backend.'
)
module = sox_backend
elif backend == 'sox_io': elif backend == 'sox_io':
module = sox_io_backend module = sox_io_backend
elif backend == 'soundfile': elif backend == 'soundfile':
if torchaudio.USE_SOUNDFILE_LEGACY_INTERFACE:
warnings.warn(
'The legacy interface of "soundfile" backend is deprecated and will be removed in 0.9.0. '
'Please migrate to the new interface.'
)
module = soundfile_backend module = soundfile_backend
else:
module = _soundfile_backend
else: else:
raise NotImplementedError(f'Unexpected backend "{backend}"') raise NotImplementedError(f'Unexpected backend "{backend}"')
...@@ -98,10 +76,8 @@ def get_audio_backend() -> Optional[str]: ...@@ -98,10 +76,8 @@ def get_audio_backend() -> Optional[str]:
""" """
if torchaudio.load == no_backend.load: if torchaudio.load == no_backend.load:
return None return None
if torchaudio.load == sox_backend.load:
return 'sox'
if torchaudio.load == sox_io_backend.load: if torchaudio.load == sox_io_backend.load:
return 'sox_io' return 'sox_io'
if torchaudio.load in [soundfile_backend.load, _soundfile_backend.load]: if torchaudio.load == soundfile_backend.load:
return 'soundfile' return 'soundfile'
raise ValueError('Unknown backend.') raise ValueError('Unknown backend.')
torchaudio/csrc/CMakeLists.txt
View file @ 33dc817c
...@@ -61,7 +61,6 @@ if (BUILD_TORCHAUDIO_PYTHON_EXTENSION) ...@@ -61,7 +61,6 @@ if (BUILD_TORCHAUDIO_PYTHON_EXTENSION)
_torchaudio _torchaudio
SHARED SHARED
pybind.cpp pybind.cpp
sox/legacy.cpp
${LIBTORCHAUDIO_SOURCES} ${LIBTORCHAUDIO_SOURCES}
) )
......
torchaudio/csrc/pybind.cpp
View file @ 33dc817c
#include <torch/extension.h> #include <torch/extension.h>
#include <torchaudio/csrc/sox/effects.h> #include <torchaudio/csrc/sox/effects.h>
#include <torchaudio/csrc/sox/io.h> #include <torchaudio/csrc/sox/io.h>
#include <torchaudio/csrc/sox/legacy.h>
PYBIND11_MODULE(_torchaudio, m) { PYBIND11_MODULE(_torchaudio, m) {
py::class_<sox_signalinfo_t>(m, "sox_signalinfo_t")
.def(py::init<>())
.def(
"__repr__",
[](const sox_signalinfo_t& self) {
std::stringstream ss;
ss << "sox_signalinfo_t {\n"
<< " rate-> " << self.rate << "\n"
<< " channels-> " << self.channels << "\n"
<< " precision-> " << self.precision << "\n"
<< " length-> " << self.length << "\n"
<< " mult-> " << self.mult << "\n"
<< "}\n";
return ss.str();
})
.def_readwrite("rate", &sox_signalinfo_t::rate)
.def_readwrite("channels", &sox_signalinfo_t::channels)
.def_readwrite("precision", &sox_signalinfo_t::precision)
.def_readwrite("length", &sox_signalinfo_t::length)
.def_readwrite("mult", &sox_signalinfo_t::mult);
py::class_<sox_encodinginfo_t>(m, "sox_encodinginfo_t")
.def(py::init<>())
.def(
"__repr__",
[](const sox_encodinginfo_t& self) {
std::stringstream ss;
ss << "sox_encodinginfo_t {\n"
<< " encoding-> " << self.encoding << "\n"
<< " bits_per_sample-> " << self.bits_per_sample << "\n"
<< " compression-> " << self.compression << "\n"
<< " reverse_bytes-> " << self.reverse_bytes << "\n"
<< " reverse_nibbles-> " << self.reverse_nibbles << "\n"
<< " reverse_bits-> " << self.reverse_bits << "\n"
<< " opposite_endian-> " << self.opposite_endian << "\n"
<< "}\n";
return ss.str();
})
.def_readwrite("encoding", &sox_encodinginfo_t::encoding)
.def_readwrite("bits_per_sample", &sox_encodinginfo_t::bits_per_sample)
.def_readwrite("compression", &sox_encodinginfo_t::compression)
.def_readwrite("reverse_bytes", &sox_encodinginfo_t::reverse_bytes)
.def_readwrite("reverse_nibbles", &sox_encodinginfo_t::reverse_nibbles)
.def_readwrite("reverse_bits", &sox_encodinginfo_t::reverse_bits)
.def_readwrite("opposite_endian", &sox_encodinginfo_t::opposite_endian);
py::enum_<sox_encoding_t>(m, "sox_encoding_t")
.value("SOX_ENCODING_UNKNOWN", sox_encoding_t::SOX_ENCODING_UNKNOWN)
.value("SOX_ENCODING_SIGN2", sox_encoding_t::SOX_ENCODING_SIGN2)
.value("SOX_ENCODING_UNSIGNED", sox_encoding_t::SOX_ENCODING_UNSIGNED)
.value("SOX_ENCODING_FLOAT", sox_encoding_t::SOX_ENCODING_FLOAT)
.value("SOX_ENCODING_FLOAT_TEXT", sox_encoding_t::SOX_ENCODING_FLOAT_TEXT)
.value("SOX_ENCODING_FLAC", sox_encoding_t::SOX_ENCODING_FLAC)
.value("SOX_ENCODING_HCOM", sox_encoding_t::SOX_ENCODING_HCOM)
.value("SOX_ENCODING_WAVPACK", sox_encoding_t::SOX_ENCODING_WAVPACK)
.value("SOX_ENCODING_WAVPACKF", sox_encoding_t::SOX_ENCODING_WAVPACKF)
.value("SOX_ENCODING_ULAW", sox_encoding_t::SOX_ENCODING_ULAW)
.value("SOX_ENCODING_ALAW", sox_encoding_t::SOX_ENCODING_ALAW)
.value("SOX_ENCODING_G721", sox_encoding_t::SOX_ENCODING_G721)
.value("SOX_ENCODING_G723", sox_encoding_t::SOX_ENCODING_G723)
.value("SOX_ENCODING_CL_ADPCM", sox_encoding_t::SOX_ENCODING_CL_ADPCM)
.value("SOX_ENCODING_CL_ADPCM16", sox_encoding_t::SOX_ENCODING_CL_ADPCM16)
.value("SOX_ENCODING_MS_ADPCM", sox_encoding_t::SOX_ENCODING_MS_ADPCM)
.value("SOX_ENCODING_IMA_ADPCM", sox_encoding_t::SOX_ENCODING_IMA_ADPCM)
.value("SOX_ENCODING_OKI_ADPCM", sox_encoding_t::SOX_ENCODING_OKI_ADPCM)
.value("SOX_ENCODING_DPCM", sox_encoding_t::SOX_ENCODING_DPCM)
.value("SOX_ENCODING_DWVW", sox_encoding_t::SOX_ENCODING_DWVW)
.value("SOX_ENCODING_DWVWN", sox_encoding_t::SOX_ENCODING_DWVWN)
.value("SOX_ENCODING_GSM", sox_encoding_t::SOX_ENCODING_GSM)
.value("SOX_ENCODING_MP3", sox_encoding_t::SOX_ENCODING_MP3)
.value("SOX_ENCODING_VORBIS", sox_encoding_t::SOX_ENCODING_VORBIS)
.value("SOX_ENCODING_AMR_WB", sox_encoding_t::SOX_ENCODING_AMR_WB)
.value("SOX_ENCODING_AMR_NB", sox_encoding_t::SOX_ENCODING_AMR_NB)
.value("SOX_ENCODING_LPC10", sox_encoding_t::SOX_ENCODING_LPC10)
//.value("SOX_ENCODING_OPUS", sox_encoding_t::SOX_ENCODING_OPUS) //
// creates a compile error
.value("SOX_ENCODINGS", sox_encoding_t::SOX_ENCODINGS)
.export_values();
py::enum_<sox_option_t>(m, "sox_option_t")
.value("sox_option_no", sox_option_t::sox_option_no)
.value("sox_option_yes", sox_option_t::sox_option_yes)
.value("sox_option_default", sox_option_t::sox_option_default)
.export_values();
py::enum_<sox_bool>(m, "sox_bool")
.value("sox_false", sox_bool::sox_false)
.value("sox_true", sox_bool::sox_true)
.export_values();
m.def(
"read_audio_file",
&torch::audio::read_audio_file,
"Reads an audio file into a tensor");
m.def(
"write_audio_file",
&torch::audio::write_audio_file,
"Writes data from a tensor into an audio file");
m.def(
"get_info",
&torch::audio::get_info,
"Gets information about an audio file");
m.def( m.def(
"get_info_fileobj", "get_info_fileobj",
&torchaudio::sox_io::get_info_fileobj, &torchaudio::sox_io::get_info_fileobj,
......
torchaudio/csrc/sox/legacy.cpp deleted 100644 → 0
View file @ 086467a1
#include <torchaudio/csrc/sox/legacy.h>
namespace torch {
namespace audio {
namespace {
/// Helper struct to safely close the sox_format_t descriptor.
struct SoxDescriptor {
explicit SoxDescriptor(sox_format_t* fd) noexcept : fd_(fd) {}
SoxDescriptor(const SoxDescriptor& other) = delete;
SoxDescriptor(SoxDescriptor&& other) = delete;
SoxDescriptor& operator=(const SoxDescriptor& other) = delete;
SoxDescriptor& operator=(SoxDescriptor&& other) = delete;
~SoxDescriptor() {
if (fd_ != nullptr) {
sox_close(fd_);
}
}
sox_format_t* operator->() noexcept {
return fd_;
}
sox_format_t* get() noexcept {
return fd_;
}
private:
sox_format_t* fd_;
};
int64_t write_audio(SoxDescriptor& fd, at::Tensor tensor) {
std::vector<sox_sample_t> buffer(tensor.numel());
AT_DISPATCH_ALL_TYPES(tensor.scalar_type(), "write_audio_buffer", [&] {
auto* data = tensor.data_ptr<scalar_t>();
std::copy(data, data + tensor.numel(), buffer.begin());
});
const auto samples_written =
sox_write(fd.get(), buffer.data(), buffer.size());
return samples_written;
}
void read_audio(SoxDescriptor& fd, at::Tensor output, int64_t buffer_length) {
std::vector<sox_sample_t> buffer(buffer_length);
int number_of_channels = fd->signal.channels;
const int64_t samples_read = sox_read(fd.get(), buffer.data(), buffer_length);
if (samples_read == 0) {
throw std::runtime_error(
"Error reading audio file: empty file or read failed in sox_read");
}
output.resize_({samples_read / number_of_channels, number_of_channels});
output = output.contiguous();
AT_DISPATCH_ALL_TYPES(output.scalar_type(), "read_audio_buffer", [&] {
auto* data = output.data_ptr<scalar_t>();
std::copy(buffer.begin(), buffer.begin() + samples_read, data);
});
}
} // namespace
std::tuple<sox_signalinfo_t, sox_encodinginfo_t> get_info(
const std::string& file_name) {
SoxDescriptor fd(sox_open_read(
file_name.c_str(),
/*signal=*/nullptr,
/*encoding=*/nullptr,
/*filetype=*/nullptr));
if (fd.get() == nullptr) {
throw std::runtime_error("Error opening audio file");
}
return std::make_tuple(fd->signal, fd->encoding);
}
int read_audio_file(
const std::string& file_name,
at::Tensor output,
bool ch_first,
int64_t nframes,
int64_t offset,
sox_signalinfo_t* si,
sox_encodinginfo_t* ei,
const char* ft) {
SoxDescriptor fd(sox_open_read(file_name.c_str(), si, ei, ft));
if (fd.get() == nullptr) {
throw std::runtime_error("Error opening audio file");
}
// signal info
const int number_of_channels = fd->signal.channels;
const int sample_rate = fd->signal.rate;
const int64_t total_length = fd->signal.length;
// multiply offset and number of frames by number of channels
offset *= number_of_channels;
nframes *= number_of_channels;
if (total_length == 0) {
throw std::runtime_error("Error reading audio file: unknown length");
}
if (offset > total_length) {
throw std::runtime_error("Offset past EOF");
}
// calculate buffer length
int64_t buffer_length = total_length;
if (offset > 0) {
buffer_length -= offset;
}
if (nframes > 0 && buffer_length > nframes) {
buffer_length = nframes;
}
// seek to offset point before reading data
if (sox_seek(fd.get(), offset, 0) == SOX_EOF) {
throw std::runtime_error(
"sox_seek reached EOF, try reducing offset or num_samples");
}
// read data and fill output tensor
read_audio(fd, output, buffer_length);
// L x C -> C x L, if desired
if (ch_first) {
output.transpose_(1, 0);
}
return sample_rate;
}
void write_audio_file(
const std::string& file_name,
const at::Tensor& tensor,
sox_signalinfo_t* si,
sox_encodinginfo_t* ei,
const char* file_type) {
if (!tensor.is_contiguous()) {
throw std::runtime_error(
"Error writing audio file: input tensor must be contiguous");
}
#if SOX_LIB_VERSION_CODE >= 918272 // >= 14.3.0
si->mult = nullptr;
#endif
SoxDescriptor fd(sox_open_write(
file_name.c_str(),
si,
ei,
file_type,
/*oob=*/nullptr,
/*overwrite=*/nullptr));
if (fd.get() == nullptr) {
throw std::runtime_error(
"Error writing audio file: could not open file for writing");
}
const auto samples_written = write_audio(fd, tensor);
if (samples_written != tensor.numel()) {
throw std::runtime_error(
"Error writing audio file: could not write entire buffer");
}
}
} // namespace audio
} // namespace torch
torchaudio/csrc/sox/legacy.h deleted 100644 → 0
View file @ 086467a1
#include <sox.h>
#include <torch/torch.h>
namespace torch {
namespace audio {
/// Reads an audio file from the given `path` into the `output` `Tensor` and
/// returns the sample rate of the audio file.
/// Throws `std::runtime_error` if the audio file could not be opened, or an
/// error occurred during reading of the audio data.
int read_audio_file(
const std::string& file_name,
at::Tensor output,
bool ch_first,
int64_t nframes,
int64_t offset,
sox_signalinfo_t* si,
sox_encodinginfo_t* ei,
const char* ft);
/// Writes the data of a `Tensor` into an audio file at the given `path`, with
/// a certain extension (e.g. `wav`or `mp3`) and sample rate.
/// Throws `std::runtime_error` when the audio file could not be opened for
/// writing, or an error occurred during writing of the audio data.
void write_audio_file(
const std::string& file_name,
const at::Tensor& tensor,
sox_signalinfo_t* si,
sox_encodinginfo_t* ei,
const char* file_type);
/// Reads an audio file from the given `path` and returns a tuple of
/// sox_signalinfo_t and sox_encodinginfo_t, which contain information about
/// the audio file such as sample rate, length, bit precision, encoding and
/// more. Throws `std::runtime_error` if the audio file could not be opened, or
/// an error occurred during reading of the audio data.
std::tuple<sox_signalinfo_t, sox_encodinginfo_t> get_info(
const std::string& file_name);
} // namespace audio
} // namespace torch
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment