Train and inference wav2vec2 using a language model

Hi everyone. Is it possible to use a language model built with KenLM for train and inference steps of Wav2Vec2 model? If so, please share any resource on this. Thanks in advance


We don’t have a lot of code that support LM decoding for Wav2Vec2 sadly.

This notebook: huggingface_notebook/xlsr_gpt.ipynb at main · voidful/huggingface_notebook · GitHub might help