專題研究week3languagemodelanddecoding-資料下載頁
2024-10-17 20:47本頁面
【導(dǎo)讀】cut-d''-f1--plement$train_text>. cut-d''-f1--plement$train_text>. -kndiscount(modifiedKneser-Ney. -text/exp/lm/. -vocab$lexicon(Lexicon,asshownon). Smoothing. thetrainingdata.Week2–LanguageModeling. Lexicon. lexicon=material/. FSA(orFSM). AnFSA“accepts”asetofstrings. circle.Note:costesafter“/”,Forfinal-. state,“2/1”meansfinal-cost1onstate2.Thisexamplemaps“ab”to(3=1+1+1).WFST. inputandoutput.Ex.Inputtape:“ac”?Outputtape:. Cost=++=. Notation:C=A。Bmeans,CisAposed. HCLG=H。C。L。G. H:HMMstructure. L:Lexicon. WhereisC?FromHCLGwehave?word. WeneedanotherWFST,U. ComposeUwithHCLG,.S=U。HCLG. Searchthebestpath(s)onSisthe. Splitthec
【正文】 WER) 22 Viterbi Decoding ? Viterbi Algorithm ?Given acoustic model and observations ?Find the best state sequence ? Best state sequence ? ? Phone sequence (AM) ? ? Word sequence (Lexicon) ? ? Best word sequence (LM) Viterbi Decoding ? ? Language model training , WFST decoding , Viterbi decoding Homework ToDo ? Step1. Finish code in and get your LM. ? Step2. Use your LM in ? . Run and (WFST decode for monophone) ? Run and (WFST decode for triphone) ? Run (Viterbi for mono) ? Run (Viterbi for triphone) ToDo (Opt.) ? Train LM : Use YOUR training text or even YOUR lexicon. ? Train LM (ngramcount) : Try different arguments. ? ? Watch online courses on coursera (Week2 LM) ? ? Read 數(shù)位語音處理概論 ? (Viterbi) ? (Language Model) ? (WFST) ? Try different AM/LM binations and report the recognition results. Questions ?
教學(xué)課件相關(guān)推薦
鄂ICP備17016276號-1