Add format override to load and related I/O functions (#1104)

test/torchaudio_unittest/sox_effect/sox_effect_test.py

@@ -8,6 +8,7 @@ from torchaudio_unittest.common_utils import (
     TempDirMixin,
     PytorchTestCase,
     skipIfNoExtension,
+    get_asset_path,
     get_sinusoid,
     get_wav_data,
     save_wav,
@@ -243,3 +244,21 @@ class TestFileFormats(TempDirMixin, PytorchTestCase):
 
         assert sr == expected_sr
         self.assertEqual(found, expected)
+
+
+@skipIfNoExtension
+class TestApplyEffectFileWithoutExtension(PytorchTestCase):
+    def test_mp3(self):
+        """Providing format allows to read mp3 without extension
+
+        libsox does not check header for mp3
+
+        https://github.com/pytorch/audio/issues/1040
+
+        The file was generated with the following command
+            ffmpeg -f lavfi -i "sine=frequency=1000:duration=5" -ar 16000 -f mp3 test_noext
+        """
+        effects = [['band', '300', '10']]
+        path = get_asset_path("mp3_without_ext")
+        _, sr = sox_effects.apply_effects_file(path, effects, format="mp3")
+        assert sr == 16000

test/torchaudio_unittest/sox_io_backend/info_test.py

@@ -167,3 +167,20 @@ class TestInfoOpus(PytorchTestCase):
         assert info.sample_rate == 48000
         assert info.num_frames == 32768
         assert info.num_channels == num_channels
+
+
+@skipIfNoExtension
+class TestLoadWithoutExtension(PytorchTestCase):
+    def test_mp3(self):
+        """Providing `format` allows to read mp3 without extension
+
+        libsox does not check header for mp3
+
+        https://github.com/pytorch/audio/issues/1040
+
+        The file was generated with the following command
+            ffmpeg -f lavfi -i "sine=frequency=1000:duration=5" -ar 16000 -f mp3 test_noext
+        """
+        path = get_asset_path("mp3_without_ext")
+        sinfo = sox_io_backend.info(path, format="mp3")
+        assert sinfo.sample_rate == 16000

test/torchaudio_unittest/sox_io_backend/load_test.py

-import os
 import itertools
 
 from torchaudio.backend import sox_io_backend
@@ -355,26 +354,18 @@ class TestLoadParams(TempDirMixin, PytorchTestCase):
         self.assertEqual(found, expected)
 
 
-@skipIfNoExec('sox')
 @skipIfNoExtension
-class TestLoadExtensionLess(TempDirMixin, PytorchTestCase):
-    """Given `format` parameter, `sox_io_backend.load` can load files without extension"""
-    original = None
-    path = None
+class TestLoadWithoutExtension(PytorchTestCase):
+    def test_mp3(self):
+        """Providing format allows to read mp3 without extension
 
-    def _make_file(self, format_):
-        sample_rate = 8000
-        path = self.get_temp_path(f'test.{format_}')
-        sox_utils.gen_audio_file(f'{path}', sample_rate, num_channels=2)
-        self.original = sox_io_backend.load(path)[0]
-        self.path = os.path.splitext(path)[0]
-        os.rename(path, self.path)
-
-    @parameterized.expand([
-        ('WAV', ), ('wav', ), ('MP3', ), ('mp3', ), ('FLAC', ), ('flac',),
-    ], name_func=name_func)
-    def test_format(self, format_):
-        """Providing format allows to read file without extension"""
-        self._make_file(format_)
-        found, _ = sox_io_backend.load(self.path)
-        self.assertEqual(found, self.original)
+        libsox does not check header for mp3
+
+        https://github.com/pytorch/audio/issues/1040
+
+        The file was generated with the following command
+            ffmpeg -f lavfi -i "sine=frequency=1000:duration=5" -ar 16000 -f mp3 test_noext
+        """
+        path = get_asset_path("mp3_without_ext")
+        _, sr = sox_io_backend.load(path, format="mp3")
+        assert sr == 16000

third_party/CMakeLists.txt

@@ -84,7 +84,6 @@ ExternalProject_Add(libsox
   DOWNLOAD_DIR ${ARCHIVE_DIR}
   URL https://downloads.sourceforge.net/project/sox/sox/14.4.2/sox-14.4.2.tar.bz2
   URL_HASH SHA256=81a6956d4330e75b5827316e44ae381e6f1e8928003c6aa45896da9041ea149c
-  PATCH_COMMAND patch -p0 < ${CMAKE_CURRENT_SOURCE_DIR}/patch/libsox.patch
   # OpenMP is by default compiled against GNU OpenMP, which conflicts with the version of OpenMP that PyTorch uses.
   # See https://github.com/pytorch/audio/pull/1026
   CONFIGURE_COMMAND ${CMAKE_CURRENT_SOURCE_DIR}/build_codec_helper.sh ${CMAKE_CURRENT_SOURCE_DIR}/src/libsox/configure ${COMMON_ARGS} --with-lame --with-flac --with-mad --with-oggvorbis --without-alsa --without-coreaudio --without-png --without-oss --without-sndfile --with-opus --with-amrwb --with-amrnb --disable-openmp

third_party/patch/libsox.patch

---- src/formats.c	2014-10-27 02:55:50.000000000 +0000
-+++ src/formats_new.c	2020-11-18 19:14:17.398689371 +0000
-@@ -91,6 +91,21 @@
- 
-   if (ext && !strcasecmp(ext, "snd"))
-   CHECK(sndr  , 7, 1, ""     , 0,  2, "\0")
-+
-+#if defined HAVE_MP3 && (defined STATIC_MP3 || !defined HAVE_LIBLTDL)
-+  // http://www.mp3-tech.org/programmer/frame_header.html
-+  // Check the first two bytes are
-+  // expected             1111_1111 111X_X01X
-+  // mask         \xEEE6 (1111_1111 1110_0110)
-+  // masked value \xEEE2 (1111_1111 1110_0010)
-+  if (len >= 2 && !memcmp(data, "\xFF", 1)) {
-+    unsigned char second_byte = data[1];
-+    unsigned char mask = 0xE6;
-+    unsigned char expected = 0xE2;
-+    if ((second_byte & mask) == expected)
-+      return "mp3";
-+  }
-+#endif
-   #undef CHECK
- 
- #if HAVE_MAGIC

torchaudio/backend/_soundfile_backend.py

@@ -45,6 +45,7 @@ def load(
     num_frames: int = -1,
     normalize: bool = True,
     channels_first: bool = True,
+    format: Optional[str] = None,
 ) -> Tuple[torch.Tensor, int]:
     """Load audio data from file.
 
@@ -99,6 +100,8 @@ def load(
         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:
         torch.Tensor:

torchaudio/backend/sox_io_backend.py

@@ -9,20 +9,27 @@ from .common import AudioMetaData
 
 
 @_mod_utils.requires_module('torchaudio._torchaudio')
-def info(filepath: str) -> AudioMetaData:
+def info(
+        filepath: str,
+        format: Optional[str] = None,
+) -> AudioMetaData:
     """Get signal information of an audio file.
 
     Args:
         filepath (str or pathlib.Path):
             Path to audio file. This function also handles ``pathlib.Path`` objects,
             but is annotated as ``str`` for TorchScript compatibility.
+        format (str, optional):
+            Override the format detection with the given format.
+            Providing the argument might help when libsox can not infer the format
+            from header or extension,
 
     Returns:
         AudioMetaData: Metadata of the given audio.
     """
     # Cast to str in case type is `pathlib.Path`
     filepath = str(filepath)
-    sinfo = torch.ops.torchaudio.sox_io_get_info(filepath)
+    sinfo = torch.ops.torchaudio.sox_io_get_info(filepath, format)
     return AudioMetaData(sinfo.get_sample_rate(), sinfo.get_num_frames(), sinfo.get_num_channels())
 
 
@@ -33,6 +40,7 @@ def load(
         num_frames: int = -1,
         normalize: bool = True,
         channels_first: bool = True,
+        format: Optional[str] = None,
 ) -> Tuple[torch.Tensor, int]:
     """Load audio data from file.
 
@@ -93,6 +101,10 @@ def load(
         channels_first (bool):
             When True, the returned Tensor has dimension ``[channel, time]``.
             Otherwise, the returned Tensor's dimension is ``[time, channel]``.
+        format (str, optional):
+            Override the format detection with the given format.
+            Providing the argument might help when libsox can not infer the format
+            from header or extension,
 
     Returns:
         torch.Tensor:
@@ -103,7 +115,7 @@ def load(
     # Cast to str in case type is `pathlib.Path`
     filepath = str(filepath)
     signal = torch.ops.torchaudio.sox_io_load_audio_file(
-        filepath, frame_offset, num_frames, normalize, channels_first)
+        filepath, frame_offset, num_frames, normalize, channels_first, format)
     return signal.get_tensor(), signal.get_sample_rate()
 
 

torchaudio/csrc/register.cpp

@@ -49,7 +49,14 @@ TORCH_LIBRARY(torchaudio, m) {
 
   m.def("torchaudio::sox_io_get_info", &torchaudio::sox_io::get_info);
   m.def(
-      "torchaudio::sox_io_load_audio_file",
+      "torchaudio::sox_io_load_audio_file("
+          "str path,"
+          "int? frame_offset=None,"
+          "int? num_frames=None,"
+          "bool? normalize=True,"
+          "bool? channels_first=False,"
+          "str? format=None"
+      ") -> __torch__.torch.classes.torchaudio.TensorSignal",
       &torchaudio::sox_io::load_audio_file);
   m.def(
       "torchaudio::sox_io_save_audio_file",

torchaudio/csrc/sox_effects.cpp

@@ -92,13 +92,14 @@ c10::intrusive_ptr<TensorSignal> apply_effects_file(
     const std::string path,
     std::vector<std::vector<std::string>> effects,
     c10::optional<bool>& normalize,
-    c10::optional<bool>& channels_first) {
+    c10::optional<bool>& channels_first,
+    c10::optional<std::string>& format) {
   // Open input file
   SoxFormat sf(sox_open_read(
       path.c_str(),
       /*signal=*/nullptr,
       /*encoding=*/nullptr,
-      /*filetype=*/nullptr));
+      /*filetype=*/format.has_value() ? format.value().c_str() : nullptr));
 
   validate_input_file(sf);
 

torchaudio/csrc/sox_effects.h

@@ -19,7 +19,8 @@ c10::intrusive_ptr<torchaudio::sox_utils::TensorSignal> apply_effects_file(
     const std::string path,
     std::vector<std::vector<std::string>> effects,
     c10::optional<bool>& normalize,
-    c10::optional<bool>& channels_first);
+    c10::optional<bool>& channels_first,
+    c10::optional<std::string>& format);
 
 } // namespace sox_effects
 } // namespace torchaudio

torchaudio/csrc/sox_io.cpp

@@ -30,12 +30,14 @@ int64_t SignalInfo::getNumFrames() const {
   return num_frames;
 }
 
-c10::intrusive_ptr<SignalInfo> get_info(const std::string& path) {
+c10::intrusive_ptr<SignalInfo> get_info(
+    const std::string& path,
+    c10::optional<std::string>& format) {
   SoxFormat sf(sox_open_read(
       path.c_str(),
       /*signal=*/nullptr,
       /*encoding=*/nullptr,
-      /*filetype=*/nullptr));
+      /*filetype=*/format.has_value() ? format.value().c_str() : nullptr));
 
   if (static_cast<sox_format_t*>(sf) == nullptr) {
     throw std::runtime_error("Error opening audio file");
@@ -52,7 +54,8 @@ c10::intrusive_ptr<TensorSignal> load_audio_file(
     c10::optional<int64_t>& frame_offset,
     c10::optional<int64_t>& num_frames,
     c10::optional<bool>& normalize,
-    c10::optional<bool>& channels_first) {
+    c10::optional<bool>& channels_first,
+    c10::optional<std::string>& format) {
   const auto offset = frame_offset.value_or(0);
   if (offset < 0) {
     throw std::runtime_error(
@@ -78,7 +81,7 @@ c10::intrusive_ptr<TensorSignal> load_audio_file(
   }
 
   return torchaudio::sox_effects::apply_effects_file(
-      path, effects, normalize, channels_first);
+      path, effects, normalize, channels_first, format);
 }
 
 void save_audio_file(

torchaudio/csrc/sox_io.h

@@ -21,14 +21,17 @@ struct SignalInfo : torch::CustomClassHolder {
   int64_t getNumFrames() const;
 };
 
-c10::intrusive_ptr<SignalInfo> get_info(const std::string& path);
+c10::intrusive_ptr<SignalInfo> get_info(
+    const std::string& path,
+    c10::optional<std::string>& format);
 
 c10::intrusive_ptr<torchaudio::sox_utils::TensorSignal> load_audio_file(
     const std::string& path,
     c10::optional<int64_t>& frame_offset,
     c10::optional<int64_t>& num_frames,
     c10::optional<bool>& normalize,
-    c10::optional<bool>& channels_first);
+    c10::optional<bool>& channels_first,
+    c10::optional<std::string>& format);
 
 void save_audio_file(
     const std::string& file_name,

torchaudio/sox_effects/sox_effects.py

-from typing import List, Tuple
+from typing import List, Tuple, Optional
 
 import torch
 
@@ -157,6 +157,7 @@ def apply_effects_file(
         effects: List[List[str]],
         normalize: bool = True,
         channels_first: bool = True,
+        format: Optional[str] = None,
 ) -> Tuple[torch.Tensor, int]:
     """Apply sox effects to the audio file and load the resulting data as Tensor
 
@@ -180,6 +181,10 @@ def apply_effects_file(
             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):
+            Override the format detection with the given format.
+            Providing the argument might help when libsox can not infer the format
+            from header or extension,
 
     Returns:
         Tuple[torch.Tensor, int]: Resulting Tensor and sample rate.
@@ -249,5 +254,6 @@ def apply_effects_file(
     """
     # Get string representation of 'path' in case Path object is passed
     path = str(path)
-    signal = torch.ops.torchaudio.sox_effects_apply_effects_file(path, effects, normalize, channels_first)
+    signal = torch.ops.torchaudio.sox_effects_apply_effects_file(
+        path, effects, normalize, channels_first, format)
     return signal.get_tensor(), signal.get_sample_rate()