I am looking for a good decoder to pair my acoustic model output with my language model output.
Since my acoustic model outputs a string of phonemes they need to be fuzzy matched with the pronunciation model in order to generate possible sentences for evaluation by the language model.
The problem is that the possibilities (even for small sentences) can become overwhelming and certainly explodes the memory on my 64GB machine.
My pipeline is mainly implemented in python and all of my models have been created by me so I do not have a convention similar to HTK or Kaldi.
I was wondering if anyone knows of an open-source alternative for homemade models. I am not too worried about speed since my project is for research and not a product but I need something efficient enough to be usable.