Is language model affect the training accuracy or prediction of the text ,and is it worth to spend much time on it

Is language model affect the training accuracy or prediction of the test text ,and is it worth to spend much time on it. because I planning to add my own dataset in libirispeech lm-norm.txt and make a lm.

Yes, it does affect training.

In what way it affects the training can you please explain the detail of language model purpose.
Thanks

The language model does not affect the training process, but it does affect the predictions during test epochs. See for example this explanation: https://medium.com/corti-ai/ctc-networks-and-language-models-prefix-beam-search-explained-c11d1ee23306

3 Likes

Thanks that really helped me a lot about CTC beam search and lm. To make own language model I have to give transcript data in kenml so that it will produce n-gram model.I have doubt what is the input data structure of kenml for example I have to send data like sentence1 \n sentence2 \n sentence3 .txt file .I tried to refer the libri norm .txt but it is too large to open so I don’t have reference of input data format and one more what is trie