Closed Teenuu closed 1 year ago
Did you get right result when decoding without LM?
Did you get right result when decoding without LM?
Yes I got the correct result without LM. I am only facing issue with my own trained model with LM decoding.
010010007 yd y yeh y yew yea y yeo 010010041 yi yea yeh yt zoe 010010064 yd y yeh yi qr yea yea yi qgh yi 010010093 yt yea zoe zoggs yields yn y halts y 010010100 yi yi yeh ypu yi yeh yea zog qr y 010010103 yeh yea yeh y 010010144 y yea zoe zoggs yea yd yi 010010231 y yeh yea yeh yeh y y yea 010010281 qr yeo yea yt yi zp yea yl yea yeo The random symbols are your modeling unit?
010010007 yd y yeh y yew yea y yeo 010010041 yi yea yeh yt zoe 010010064 yd y yeh yi qr yea yea yi qgh yi 010010093 yt yea zoe zoggs yields yn y halts y 010010100 yi yi yeh ypu yi yeh yea zog qr y 010010103 yeh yea yeh y 010010144 y yea zoe zoggs yea yd yi 010010231 y yeh yea yeh yeh y y yea 010010281 qr yeo yea yt yi zp yea yl yea yeo The random symbols are your modeling unit?
They are not part of model units. These words are there in my lexicon.
I think it's same problem with it. https://github.com/wenet-e2e/wenet/issues/1673
please follow https://github.com/wenet-e2e/wenet/issues/1673
Describe the bug Decoding with WFST LM is working fine with pre-trained models. But with my own model getting random output using LM decoding. Sample out is given below. Please tell what might be the possible issues. uttid Transcription 010010007 yd y yeh y yew yea y yeo 010010041 yi yea yeh yt zoe 010010064 yd y yeh yi qr yea yea yi qgh yi 010010093 yt yea zoe zoggs yields yn y halts y 010010100 yi yi yeh ypu yi yeh yea zog qr y 010010103 yeh yea yeh y 010010144 y yea zoe zoggs yea yd yi 010010231 y yeh yea yeh yeh y y yea 010010281 qr yeo yea yt yi zp yea yl yea yeo ~
To Reproduce Steps to reproduce the behavior: