For automatic speech recognition models, ([Wav2Vec2ForCTC]), the model expects a tensor of dimension (batch_size,
  target_length) with each value corresponding to the expected label of each individual token.