How to import a custom dataset to fine tune wav2vec


I’m seeking for your advice to fine-tune wav2vec with my custom dataset

I have tried to follow these this tutorials : Fine-tuning XLSR-Wav2Vec2 for Multi-Lingual ASR with :hugs: Transformers

and I made my custom dataset as a pandas dataframe contains 3 coulmns:

  • first column have audi file name like “audio1.wav”
  • second column have the transcriptions like “hello world”
  • third coulmn have the path for the audio file like “./data/audio1.wav”

and then make it as dataset using

from datasets import Dataset, load_metric

train_data = Dataset.from_pandas(train_df)
test_data = Dataset.from_pandas(test_df)

and i did all the steps and trained the model also, but the problem is there is no prediction coming out, i got only blank every time i try to make prediction

so please help me how to solve this problem, is there is a better way to make custom dataset.