🚀 Feature request

Implementing Conformer version of context encoder for Wav2Vec2.

Motivation

Fairseq has recently updated their Wav2Vec2 codebase, incorporating Conformer-based encoder as an option with already existing Transformer-based version.

As long as a decent number of the new SOTA-matching papers are utilizing Conformers for either of ctc or transducer models, it would be a great addition to Transformers library to have an option of using transformer or conformer based encoder for Wav2Vec2ForCTC

Some of the papers: JUST, Pushing the Limits of Semi-Supervised Learning for ASR