(Audio classification pipeline) ValueError: ffmpeg was not found but is required to load audio files from filename

Beginners
1

Greetings Huggingface community!

I have been following the examples in the docs, for the example of audio pipeline under the ‘Pipelines for inference’ tutorial, I tried out the follwing example:

from transformers import pipeline
audio_classifier = pipeline(task='audio-classification',model = \
                            "ehcalabres/wav2vec2-lg-xlsr-en-speech-emotion-recognition")
audio_classifier("1.wav")  # A .wav file stored locally in the same directory

I get the following error

ValueError: ffmpeg was not found but is required to load audio files from filename

I’m running all of this under a conda environment and made sure that ffmpeg was installed.

Any assistance would be greatly appreciated

1 Like
2

I have the same issue, any update on that?

3

For anyone encountering this issue, I found a solution. Apparently python ffmpeg-python package doesn’t install ffmpeg itself so you have to install it manually on your machine Download FFmpeg

4

@Avienir Hey, I am facing the same issue in Windows 10. So I downsload the ffmpeg and where should I place the files? Could you please help me?

5

You can use choco in windows10，if you have，I use choco install ffmpeg solve the problem.

7

Installing ffmpeg also worked for me. I’m a mac user, and used brew to install ffmpeg.

$ brew install ffmpeg

2 Likes
8

I fixed it in Windows 11 by following steps:

  1. Uninstall ffmpeg
  • pip3 uninstall ffmpeg
  • pip3 uninstall ffmpeg-python
  • pip uninstall ffmpeg
  • pip uninstall ffmpeg-python
  • brew uninstall ffmpeg

  1. Install ffmpeg binary using this guide.

  2. Install ffmpeg for python

  • pip3 install ffmpeg
  • pip3 install ffmpeg-python
  • pip install ffmpeg
  • pip install ffmpeg-python

That’s all it should work :slight_smile:

11 Likes
9

Also useful on win10

10

I’m on Visual Studio Code environment, and I have the same issue.
I did all the steps asked by jb-infivr, but I have the same error.
Thank you very much for the work :muscle:

Edit :
I tried to modify the code using this :
display(classifier(minds[0]['audio']['array']))

I don’t have errors anymore, but the output seems to not have a good answer :

[{'score': 0.9950760006904602, 'label': 'app_error'},
 {'score': 0.0014388954732567072, 'label': 'address'},
 {'score': 0.00135265220887959, 'label': 'high_value_payment'},
 {'score': 0.000736881687771529, 'label': 'card_issues'},
 {'score': 0.0003506052598822862, 'label': 'abroad'}]

The expected result :

[
    {"score": 0.9631525278091431, "label": "pay_bill"},
    {"score": 0.02819698303937912, "label": "freeze"},
    {"score": 0.0032787492964416742, "label": "card_issues"},
    {"score": 0.0019414445850998163, "label": "abroad"},
    {"score": 0.0008378693601116538, "label": "high_value_payment"},
]

If you have any idea, I’m interested !

Thank you :slight_smile:

11

I am also facing the same issue. Apparently ffmpeg is used for audio compression (?). Is there anyway where we can pass on the uncompressed audio file in the pipeline. My aim is to use it without ffmpeg being installed in the system. Faster-whisper package runs without ffmpeg in the system.
Any other suggestions are also appreciated.
Thanks in advance

12

conda install ffmpeg solved it for me!

3 Likes
13

this is very useful :blush:

1 Like