Can't load feature extractor for 'facebook/wav2vec2-large-xlsr-53'

Hi @Gastron, TParcollet, I was very happy coming across speechbrain toolkit for my master project on African ASR, However i have been having serious challenges for over 24 hours trying to train a model using FONGBE dataset which i have download and is supported by speechbrain network. Below is the actual error i’m having.

f"Can't load feature extractor for '{pretrained_model_name_or_path}'. If you were trying to load it "
OSError: Can't load feature extractor for 'facebook/wav2vec2-large-xlsr-53'. If you were trying to load it from 'https://huggingface.co/models', make sure you don't have a local directory with the same name. Otherwise, make sure 'facebook/wav2vec2-large-xlsr-53' is the correct path to a directory containing a preprocessor_config.json file

when i run this command to start training python train_with_wav2vec2.py hparams/train_fon_with_wav2vec.yaml --data_folder=/localscratch/ALFFA_PUBLIC/ASR/FONGBE/data/

Acoording to the documentation on the speechbrain huggingface page.

my yaml conifguration is this

# Url for xlsr wav2vec2
#wav2vec2_hub: C:\Users\nathk\Documents\TacoProject\brain\speechbrain\xlsr_53_56k.pt
wav2vec2_hub: facebook/wav2vec2-large-xlsr-53
#wav2vec2_hub: C:/Users/nathk/Documents/TacoProject/brain/speechbrain/recipes/DVoice/ASR/CTC/results/wav2vec2_ctc_FONGBE/1250/save/wav2vec2_checkpoint/config.json


# Data files
data_folder: /localscratch/ALFFA_PUBLIC/ASR/FONGBE/data
train_csv_file: !ref <data_folder>/train.csv

I have manually downloaded the wav2vec2-large-xlsr-53 model and then replace the path in yaml file in this manner: wav2vec2_hub: C:\Users\nathk\Documents\TacoProject\brain\speechbrain\xlsr_53_56k.pt But still nothing seems to work.

Please can you assist me. FONGBE documentation link which i’m using : https://huggingface.co/speechbrain/asr-wav2vec2-dvoice-fongbe

[TParcollet]

Yes, but make sure that it contains ALL the files from the repo 😃

[Adel-Moumen]

Hello,

it seems there has been no activity for a long time, so I am closing this issue.

Feel free to reopen it if you still encounter the problem. Thank you. 🙂