HiddenMarkovModel隐马尔可夫模型-国立云林科技大学.PPTVIP

Intelligent Database Systems Lab N.Y.U.S.T. I. M. Intelligent Database Systems Lab 國立雲林科技大學 National Yunlin University of Science and Technology * An Approach to Using the Web as a Live Corpus  for Spoken Transliteration Name Access Presenter : You Lin Chen Authors Ming-Shun Lin , Chia-Ping Chen and Hsin-Hsi Chen 2006.CLCLP * Outline Motivation Objective Methodology Experiments Conclusion Comments Motivation Recognizing transliteration names is challenging due to their flexible formulation and lexical coverage. * Objectives In our approach, we employ the Web as a giant corpus. The patterns extracted from the Web are used as a live dictionary to correct speech recognition errors. For collecting transliteration names, we propose a named entity (NE) ontology generation engine, called the XNE-Tree engine. * Methodology * Hidden Markov Model 隱馬爾可夫模型 隱馬爾可夫模型狀態變遷圖 x — 隱含狀態 y — 可觀察的輸出 a — 轉換機率（transition probabilities） b — 輸出機率（output probabilities） * 「賭場用骰子」的狀態轉換圖 HMM複雜度 假定有N個音節標記，給定的詞串中有M個單字，考慮最壞的情況下，每個單字都有N個可能的音節出現，則可能的路徑有NM條，隨著M的增加，需要計算的可能路徑數目以指數模式增長，即演算法時間複雜性成指數成長。 p q n v N p q n v p q n v p q n v p q n v p q n v M …… Viterbi Viterbi演算法是dynamic programing的一種 假設有N種音節，在長度為M的字串中 遞迴計算並儲存每個單字中每種音節的最佳到達路徑與機率值 當到達最後一個單字時，計算最佳的字串 由最後一個單字的最佳字串所儲存的路徑，依序回推前一個的最佳單字，如此便能找到最佳解 p q n v p q n v 每一個單字的音節狀態裡皆紀錄著最佳到達的路徑△及最佳路徑的機率δ m m+1 aij Patricia tree * PAT tree is a data structure that allows for a particularly fast implementation of many important string operations. 優點：不依賴詞庫或文法就能夠擷取中文字，且能夠處理較長、科技名詞等特殊字詞。 Methodology * Distance(塔莫克魯斯 , 湯姆克魯斯)=2 Ontology * Named entity ontology generation process * * Experiments * Conclusion In this study, we employ the Web as a giant corpus to correct transliteration name recognition errors. For collecting transliteration names beforehand, we propose using a named entity (NE) ontology generation engine. we conclude that the error recovery procedure and NE ontology can helpful to the ASR model. * Comments Advantage … D