Does score in beam dump not show all score for select best hypothesis ?

[trangtv57]

Question

Hi, I have an utterance where i have oov "titan" it's can be same as speaking 2 word in dictionary {"ti" , "tan" }. In lexicon: titan _ti _tan I am using beam dump for get score of each beamsearch, as sample below. Score of AM is same in 3 beam, although score of LM beam third is biggest, but combine LM and AM score final score of third is lowest. So I wonder the score show in beam dump is full or not, how it's happen?

Sample beam dump: id1 | 1705.429154 | 1721.264524 | -43.264816 | 18.750000 | rồi ốc titan a nói chung ốc ti tan rồi nè rồi nói chung là đồng hồ id2 | 1705.412166 | 1721.264524 | -44.503370 | 31.250000 | rồi ốc ti tan a nói chung ốc ti tan rồi nè rồi nói chung là đồng hồ id3 | 1705.152164 | 1721.264524 | -42.446229 | 6.250000 | rồi ốc titan a nói chung ốc titan rồi nè rồi nói chung là đồng hồ

thanks you.

[tlikhomanenko]

The first score is total score which is equal to AM score + LM score + silscore #silences + wordscore #words. Do you have silscore/wordscore non-zero?

[trangtv57]

hi @tlikhomanenko , I miss understand about silsscore, and word score. Then i can figure out how its happen thank you.