基于多詞匯樹(shù)的對(duì)話語(yǔ)音識(shí)別搜索策略

在查詢(xún)類(lèi)對(duì)話語(yǔ)音識(shí)別系統(tǒng)中，詞匯中的少數(shù)幾類(lèi)詞往往占有詞匯量的很大部分，利
用這一重要特點(diǎn)，可以對(duì)傳統(tǒng)的基于單樹(shù)結(jié)構(gòu)的搜索算法進(jìn)行改進(jìn)。本文在采用基于類(lèi)的語(yǔ)言模型的條件下，在查詢(xún)類(lèi)對(duì)話語(yǔ)音識(shí)別系統(tǒng)中，對(duì)詞匯中的若干大類(lèi)單獨(dú)建樹(shù)，從而免除了多數(shù)詞匯的語(yǔ)言模型look-ahead 計(jì)算時(shí)間和存儲(chǔ)空間，并實(shí)現(xiàn)了語(yǔ)言分的提前精確運(yùn)用。針對(duì)機(jī)票預(yù)訂口語(yǔ)對(duì)話系統(tǒng)的實(shí)驗(yàn)表明，該搜索策略與單樹(shù)方法相比，在同樣的計(jì)算資源條件下，時(shí)間效率有顯著提高。
關(guān)鍵詞: 語(yǔ)音識(shí)別 對(duì)話系統(tǒng) 搜索 詞樹(shù) look-ahead

Abstract A few of classes often occupy large proportion of vocabulary in speech recognition for many information query dialog system, which can be used to improve traditional search algorithm based on single lexicon tree. Under the condition of using class based n-gram language model, this paper presents a new search strategy which builds separate trees for some big word classes ( classes that own many words) to dispenses with look-ahead computation of many words in decoding process and to use precise probabilities of language model earlier. The experiments on flight reservation dialog system show that the time efficiency of the presented method is obviously higher than single-tree method under equal computational cost.
Key Words speech recognition; dialog system; search; lexicon tree; look-ahead