Inference with Language Model
#3
by 5hr3ya5 - opened
I tried to do inference of model with language model (available in the repo) as per the process given by:
https://huggingface.co/blog/wav2vec2-with-ngram
The process does not show any error.
However, the inference results are significantly degraded compared to inference without language model.
Even the output transcripts do not reflect any of the unigrams used for language model training, which is strange ! Any solution ?