🚀 Feature request

This is a feature request to add Wav2Vec2 Pretraining functionality to the transformers library. This is a "Good Second Issue" feature request, which means that interested contributors should have some experience with the transformers library and ideally also with training/fine-tuning Wav2Vec2.

Motivation

The popular [Wav2Vec2](https://huggingface.co/models?filter=w

Is there a way to train a bidirectional RNN (like LSTM or GRU) on trax nowadays?

From paper, it mentioned

Instead, the training data generator chooses 15% of tokens at random, e.g., in the sentence my
dog is hairy it chooses hairy.

It means that 15% of token will be choose for sure.

From https://github.com/codertimo/BERT-pytorch/blob/master/bert_pytorch/dataset/dataset.py#L68,
for every single token, it has 15% of chance that go though the followup procedure.

Till now, we have released the English model which is trained on
academic datasets. We are planning to support recognition models of more languages.

If you want to support a new language, please provide us with two files:

1. A char_list.txt file, which lists all the characters that used in the new language.
2. A dict_list.txt file, which lists the words in the new language as more as po

Hi, I am so interesting in your project, and wonder if you need contributor and how could I make my own contribution?