Torchaudio installed through PIP not detecting ffmpeg from conda-forge channel

[Kavinraj-N]

🐛 Describe the bug

torchaudio not detecting ffmpeg installed from the conda-forge channel.

Environment:

To reproduce, set up the following environment:

conda create -n test_env python=3.10.6 ffmpeg=6.1.2 -c conda-forge -y
conda activate test_env

pip install torch==2.2.1+cpu -f https://download.pytorch.org/whl/torch_stable.html
pip install torchaudio==2.2.1+cpu -f https://download.pytorch.org/whl/torch_stable.html

Code to reproduce:

import torchaudio
import torio

print(torchaudio.list_audio_backends())
>>> []
print(torio.utils.ffmpeg_utils.get_versions())
>>>
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/kavinraj.nac/miniconda3/envs/test_env/lib/python3.10/site-packages/torio/utils/ffmpeg_utils.py", line 19, in get_versions
    return ffmpeg_ext.get_versions()
  File "/home/kavinraj.nac/miniconda3/envs/test_env/lib/python3.10/site-packages/torio/_extension/utils.py", line 25, in __getattr__
    self._import_once()
  File "/home/kavinraj.nac/miniconda3/envs/test_env/lib/python3.10/site-packages/torio/_extension/utils.py", line 39, in _import_once
    self.module = self.import_func()
  File "/home/kavinraj.nac/miniconda3/envs/test_env/lib/python3.10/site-packages/torio/_extension/utils.py", line 143, in _init_ffmpeg
    ext = _find_ffmpeg_extension(ffmpeg_vers)
  File "/home/kavinraj.nac/miniconda3/envs/test_env/lib/python3.10/site-packages/torio/_extension/utils.py", line 122, in _find_ffmpeg_extension
    raise ImportError(
ImportError: Failed to intialize FFmpeg extension. Tried versions: ['6', '5', '4', '']. Enable DEBUG logging to see more details about the error.
>>> 

Versions

PyTorch version: 2.2.1+cpu
Is debug build: False
CUDA used to build PyTorch: Could not collect
ROCM used to build PyTorch: N/A

OS: Ubuntu 20.04.2 LTS (x86_64)
GCC version: (Ubuntu 9.4.0-1ubuntu1~20.04.2) 9.4.0
Clang version: Could not collect
CMake version: Could not collect
Libc version: glibc-2.31

Python version: 3.10.6 | packaged by conda-forge | (main, Aug 22 2022, 20:36:39) [GCC 10.4.0] (64-bit runtime)
Python platform: Linux-5.15.0-130-generic-x86_64-with-glibc2.31
Is CUDA available: False
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to: N/A
GPU models and configuration:
GPU 0: NVIDIA RTX A6000
GPU 1: NVIDIA RTX A6000
GPU 2: NVIDIA RTX A6000
GPU 3: NVIDIA RTX A6000
GPU 4: NVIDIA RTX A6000
GPU 5: NVIDIA RTX A6000
GPU 6: NVIDIA RTX A6000
GPU 7: NVIDIA RTX A6000

Nvidia driver version: 535.183.01
cuDNN version: Could not collect
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

CPU:
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
Address sizes: 43 bits physical, 48 bits virtual
CPU(s): 256
On-line CPU(s) list: 0-255
Thread(s) per core: 2
Core(s) per socket: 64
Socket(s): 2
NUMA node(s): 2
Vendor ID: AuthenticAMD
CPU family: 23
Model: 49
Model name: AMD EPYC 7662 64-Core Processor
Stepping: 0
Frequency boost: enabled
CPU MHz: 3295.616
CPU max MHz: 2154.2959
CPU min MHz: 1500.0000
BogoMIPS: 4000.00
Virtualization: AMD-V
L1d cache: 4 MiB
L1i cache: 4 MiB
L2 cache: 64 MiB
L3 cache: 512 MiB
NUMA node0 CPU(s): 0-63,128-191
NUMA node1 CPU(s): 64-127,192-255
Vulnerability Gather data sampling: Not affected
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Not affected
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed: Mitigation; untrained return thunk; SMT enabled with STIBP protection
Vulnerability Spec rstack overflow: Mitigation; safe RET
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl and seccomp
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Retpolines; IBPB conditional; STIBP always-on; RSB filling; PBRSB-eIBRS Not affected; BHI Not affected
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 cqm rdt_a rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local clzero irperf xsaveerptr rdpru wbnoinvd amd_ppin arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl umip rdpid overflow_recov succor smca sme sev sev_es

Versions of relevant libraries:
[pip3] torch==2.2.1+cpu
[pip3] torchaudio==2.2.1+cpu
[conda] libopenvino-pytorch-frontend 2024.4.0 h5888daf_2 conda-forge
[conda] torch 2.2.1+cpu pypi_0 pypi
[conda] torchaudio 2.2.1+cpu pypi_0 pypi

[khmyznikov]

Same for non-conda, under Windows environment. I'm trying to launch this tutorial: https://github.com/pytorch/audio/blob/main/examples/tutorials/audio_data_augmentation_tutorial.py

(env_torch) PS X:\GitHub\Windows-DevRel\Tools\ML-tests> python .\tests\workflow\audio_data_augmentation_tutorial.py
2.6.0+cpu
2.6.0+cpu
Traceback (most recent call last):
  File "X:\GitHub\Windows-DevRel\Tools\ML-tests\tests\workflow\audio_data_augmentation_tutorial.py", line 74, in <module>    waveform2 = apply_effect(waveform1, sample_rate, effect)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "X:\GitHub\Windows-DevRel\Tools\ML-tests\tests\workflow\audio_data_augmentation_tutorial.py", line 71, in apply_effect
    return effector.apply(waveform, sample_rate)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "X:\GitHub\Windows-DevRel\Tools\ML-tests\.temp\env_torch\Lib\site-packages\torchaudio\io\_effector.py", line 313, in apply
    reader = self._get_reader(waveform, sample_rate, output_sample_rate)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "X:\GitHub\Windows-DevRel\Tools\ML-tests\.temp\env_torch\Lib\site-packages\torchaudio\io\_effector.py", line 275, in _get_reader
    src = _encode(waveform, sample_rate, self.effect, muxer, encoder, self.codec_config)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "X:\GitHub\Windows-DevRel\Tools\ML-tests\.temp\env_torch\Lib\site-packages\torchaudio\io\_effector.py", line 98, in _encode
    writer = StreamWriter(buffer, format=muxer)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "X:\GitHub\Windows-DevRel\Tools\ML-tests\.temp\env_torch\Lib\site-packages\torio\io\_streaming_media_encoder.py", line 197, in __init__
    self._s = ffmpeg_ext.StreamingMediaEncoderFileObj(dst, format, buffer_size)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "X:\GitHub\Windows-DevRel\Tools\ML-tests\.temp\env_torch\Lib\site-packages\torio\_extension\utils.py", line 26, in __getattr__
    self._import_once()
  File "X:\GitHub\Windows-DevRel\Tools\ML-tests\.temp\env_torch\Lib\site-packages\torio\_extension\utils.py", line 40, in _import_once
    self.module = self.import_func()
                  ^^^^^^^^^^^^^^^^^^
  File "X:\GitHub\Windows-DevRel\Tools\ML-tests\.temp\env_torch\Lib\site-packages\torio\_extension\utils.py", line 144, in _init_ffmpeg
    ext = _find_ffmpeg_extension(ffmpeg_vers)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "X:\GitHub\Windows-DevRel\Tools\ML-tests\.temp\env_torch\Lib\site-packages\torio\_extension\utils.py", line 123, in _find_ffmpeg_extension
    raise ImportError(
ImportError: Failed to intialize FFmpeg extension. Tried versions: ['6', '5', '4', '']. Enable DEBUG logging to see more details about the error.

Tried 7-6-5 versions, PATH variable is correct.

[mattpitkin]

I also see this. E.g., running

from torio.utils.ffmpeg_utils import get_audio_encoders
get_audio_encoders()

---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
Cell In[13], line 1
----> 1 get_audio_encoders()

File ~/miniconda3/envs/tadl/lib/python3.10/site-packages/torio/utils/ffmpeg_utils.py:135, in get_audio_encoders()
    119 def get_audio_encoders() -> Dict[str, str]:
    120     """Get the available audio encoders.
    121
    122     Returns:
   (...)
    133         ... alac: ALAC (Apple Lossless Audio Codec)
    134     """
--> 135     return ffmpeg_ext.get_audio_encoders()

File ~/miniconda3/envs/tadl/lib/python3.10/site-packages/torio/_extension/utils.py:25, in _LazyImporter.__getattr__(self, item)
     24 def __getattr__(self, item):
---> 25     self._import_once()
     26     return getattr(self.module, item)

File ~/miniconda3/envs/tadl/lib/python3.10/site-packages/torio/_extension/utils.py:39, in _LazyImporter._import_once(self)
     37 def _import_once(self):
     38     if self.module is None:
---> 39         self.module = self.import_func()
     40         # Note:
     41         # By attaching the module attributes to self,
     42         # module attributes are directly accessible.
     43         # This allows to avoid calling __getattr__ for every attribute access.
     44         self.__dict__.update(self.module.__dict__)

File ~/miniconda3/envs/tadl/lib/python3.10/site-packages/torio/_extension/utils.py:143, in _init_ffmpeg()
    141 def _init_ffmpeg():
    142     ffmpeg_vers = _get_ffmpeg_versions()
--> 143     ext = _find_ffmpeg_extension(ffmpeg_vers)
    144     ext.init()
    145     if ext.get_log_level() > 8:

File ~/miniconda3/envs/tadl/lib/python3.10/site-packages/torio/_extension/utils.py:122, in _find_ffmpeg_extension(ffmpeg_vers)
    120         _LG.debug("Failed to load FFmpeg%s extension.", ffmpeg_ver, exc_info=True)
    121         continue
--> 122 raise ImportError(
    123     f"Failed to intialize FFmpeg extension. Tried versions: {ffmpeg_vers}. "
    124     "Enable DEBUG logging to see more details about the error."
    125 )

ImportError: Failed to intialize FFmpeg extension. Tried versions: ['6', '5', '4', '']. Enable DEBUG logging to see more details about the error.

I get the same errors are others reported.

But delving further, to find the source of this, if I do:

from torio._extension.utils import _load_lib

version = 6
lib = f"libtorio_ffmpeg{version}"
_load_lib(lib)

---------------------------------------------------------------------------
OSError                                   Traceback (most recent call last)
Cell In[21], line 1
----> 1 _load_lib(lib)

File ~/miniconda3/envs/tadl/lib/python3.10/site-packages/torio/_extension/utils.py:94, in _load_lib(lib)
     92 if not path.exists():
     93     return False
---> 94 torch.ops.load_library(path)
     95 return True

File ~/miniconda3/envs/tadl/lib/python3.10/site-packages/torch/_ops.py:1357, in _Ops.load_library(self, path)
   1352 path = _utils_internal.resolve_library_path(path)
   1353 with dl_open_guard():
   1354     # Import the shared library into the process, thus running its
   1355     # static (global) initialization code in order to register custom
   1356     # operators with the JIT.
-> 1357     ctypes.CDLL(path)
   1358 self.loaded_libraries.add(path)

File ~/miniconda3/envs/tadl/lib/python3.10/ctypes/__init__.py:374, in CDLL.__init__(self, name, mode, handle, use_errno, use_last_error, winmode)
    371 self._FuncPtr = _FuncPtr
    373 if handle is None:
--> 374     self._handle = _dlopen(self._name, mode)
    375 else:
    376     self._handle = handle

OSError: /home/matthew/miniconda3/envs/tadl/bin/../lib/libstdc++.so.6: version `GLIBCXX_3.4.32' not found (required by /lib/x86_64-linux-gnu/libavfilter.so.9)

It seems to be a C++ library version/linkage issue, with the library in my conda environment.

I partially solved this error by using the answer from here, i.e., installing:

conda install -c conda-forge libstdcxx-ng

although I this leads to a different GCC library error:

---------------------------------------------------------------------------
OSError                                   Traceback (most recent call last)
Cell In[6], line 1
----> 1 _load_lib(lib)

File ~/miniconda3/envs/tadl/lib/python3.10/site-packages/torio/_extension/utils.py:94, in _load_lib(lib)
     92 if not path.exists():
     93     return False
---> 94 torch.ops.load_library(path)
     95 return True

File ~/miniconda3/envs/tadl/lib/python3.10/site-packages/torch/_ops.py:1357, in _Ops.load_library(self, path)
   1352 path = _utils_internal.resolve_library_path(path)
   1353 with dl_open_guard():
   1354     # Import the shared library into the process, thus running its
   1355     # static (global) initialization code in order to register custom
   1356     # operators with the JIT.
-> 1357     ctypes.CDLL(path)
   1358 self.loaded_libraries.add(path)

File ~/miniconda3/envs/tadl/lib/python3.10/ctypes/__init__.py:374, in CDLL.__init__(self, name, mode, handle, use_errno, use_last_error, winmode)
    371 self._FuncPtr = _FuncPtr
    373 if handle is None:
--> 374     self._handle = _dlopen(self._name, mode)
    375 else:
    376     self._handle = handle

OSError: /home/matthew/miniconda3/envs/tadl/bin/../lib/libgcc_s.so.1: version `GCC_12.0.0' not found (required by /lib/x86_64-linux-gnu/libhwy.so.1)

I solved this one by linking my system libgcc_s.so.1 to the library location in my conda environment (like this suggestion) with:

$ ln -sf /usr/lib/x86_64-linux-gnu/libgcc_s.so.1 ${CONDA_PREFIX}/lib

Now running:

from torio.utils.ffmpeg_utils import get_audio_encoders
get_audio_encoders()

works!

I don't know how to do the same under Windows/for non-conda environment setups.

[sky121666]

https://blog.csdn.net/qq_34034975/article/details/146553230