Torchaudio.load is not able to find the backend soundfile

audio
1

I’m trying to read an audio file
data,sr = torchaudio.load(sample_wav)
But it is throwing an error
“RuntimeError: torchaudio.backend.soundfile_backend.load requires soundfile”

But
str(torchaudio.get_audio_backend()) → gives
‘soundfile’

torch 1.10.2
WIndows (So can’t use sox)

2

Hi @Rakshith_V, have you pip installed soundfile in your environment?

1 Like
3

ya I did
I’m able to run
import soundfile

but I have issues with reading from torchaudio

4

I am having the exact same issue on an Ubuntu image created with Docker.
DISTRIB_ID=Ubuntu
DISTRIB_RELEASE=20.04
DISTRIB_CODENAME=focal
DISTRIB_DESCRIPTION=“Ubuntu 20.04.5 LTS”

python3 test_cuda.py
/usr/local/lib/python3.8/dist-packages/torchaudio/_internal/module_utils.py:99: UserWarning: Failed to import soundfile. ‘soundfile’ backend is not available.

warnings.warn(“Failed to import soundfile. ‘soundfile’ backend is not available.”)

      PyTorch version: 1.11.0+cu113
  PyTorch can use GPU: True
   Torch CUDA version: 11.3

   Torchaudio version: 0.11.0+cu113
   Torchaudio backend: sox_io

Trochaudio backends list: [‘soundfile’, ‘sox_io’]

  Torchvision version: 0.12.0+cu113

Torchvision video backend: pyav

   Torch CUDA available: True
      CUDA device count: 1
    Current CUDA device: 0
            Device name: NVIDIA GeForce RTX 4090

pip list | grep -i sound
SoundFile 0.10.2
pip list | grep librosa
librosa 0.9.2

apt list | grep libsnd

libsndfile1-dev/focal-updates,focal-security 1.0.28-7ubuntu0.1 amd64
libsndfile1/focal-updates,focal-security 1.0.28-7ubuntu0.1 amd64
libsndifsdl2-dev/focal 0.8.3-2 amd64
libsndio-dev/focal 1.5.0-3 amd64
libsndio7.0/focal 1.5.0-3 amd64
libsndobj-dev/focal 2.6.7+ds1-3 amd64
libsndobj2v5/focal 2.6.7+ds1-3 amd64

I am also running apt update at the very top of my Dockerfile. I also started the container and just to be sure ran apt update and apt-get update and same problem.

python3 -c “import soundfile”
Traceback (most recent call last):
File “”, line 1, in
File “/usr/local/lib/python3.8/dist-packages/soundfile.py”, line 142, in
raise OSError(‘sndfile library not found’)
OSError: sndfile library not found

Now from the reading I have done online, Linux uses sox and not soundfile so why is torchaudio trying to access this in the first place?

5

The same issue occurred to me in windows 10 after installing soundfile. Simply restarting the computer fixed the issue. The environment variables seems to be missing after a fresh installation with pip install soundfile.