Can provide ASR performance results for different language models?

[Garry-sh]

Hi,
In speech tasks, i think the performance of the language model should be correlated with the ASR performance, but There are few ASR performance comparisons between different language models in many projects. Can you provide some experimental results to help verify experiment? such as transformer xl ,cnn...

Thanks for any hints!