Spoken Dialog Based Automobile Information System

• Main Authors: Shin-Cheng Ke, 柯欣成
• Other Authors: Jhing-Fa Wang
• Format: Others
• Language: zh-TW
• Published: 2001
• Online Access: http://ndltd.ncl.edu.tw/handle/89913284805839373174

碩士 === 國立成功大學 === 電機工程學系 === 89 === A mobile Information System (MIS) can provide driver many kinds of information, such as the location of gasoline station, traffic situation and shortest path to a destination. Most current MIS use screen and touch pad as communicate interface. However, this model is not convenient and safe while driving. The better interface will use spontaneous speech to communicate with the system. In this thesis, common errors of Spoken Dialog System (SDS) are first analyzed. Then, many practicable strategies are proposed to avoid and recover from these errors. These strategies focus on three components of the SDS, i.e. user, recognizer, and dialog manager. To overcome the defect of traditional guided dialog strategy that spend many turns to complete a dialog, this thesis proposes the Semi-Guided Dialog System (SGDS) to reduce the vocabulary size and shorten the dialog turns. The method is to classify all landmarks by analyzing landmark literally. Words with the same components will be classified into the same category. When recognizing, a two-pass recognition technique is adopted. The first pass is to recognize the categories, and the second pass then recognizes landmarks in the recognized categories. Using this method, the dialog turns is reduced to be one or two, and greatly improve the efficiency of the interaction. For the experiments, the keyword recognition rate arises 5% when SGDS is applied for large landmark retrieval. In addition, about 12% of recognition rate improvement is gained by applying proposed strategies of SDS error recovery. Finally, we test the dialog’s finish rate, and is greater then 89%. 圖目錄 iii 表目錄 v 第一章 緒論 1 1.1 研究背景 1 1.2 研究動機與目的 2 1.3 研究方法與步驟 3 1.4 系統架構 4 1.5 章節概要 5 第二章 語音對話系統之錯誤分析及可行的解決策略 1 2.1 語音對話系統之錯誤定義 1 2.2 各子系統造成的錯誤分析與回復策略 1 2.2.1 使用者之輸入系統無法處理 2 2.2.2 語音辨識錯誤 5 2.2.3 對話管理系統錯誤 8 第三章 半導引式對話系統 13 3.1 簡介 13 3.2 傳統導引式對話架構查詢流程 13 3.3 半導引式對話系統 16 3.3.1 漸次擷取使用者輸入資訊 17 3.3.2 地標詞表分類 17 3.3.3 內部多重辨識 18 3.4 半導引式對話系統建立流程 20 3.4.1 地標特性分析 20 3.4.2 地標詞表分類建立流程 21 3.4.3 半導引式對話系統地標查詢流程 25 第四章 對話行動資訊系統 27 4.1 對話系統之開發 27 4.1.1 資料收集模組 28 4.1.2 構句模組 34 4.1.3 語意理解模組 34 4.1.4 資料查詢模組 38 4.1.5 對話管理模組 38 4.1.6 文字翻語音模組 41 4.2 行動資訊系統架構 42 4.3 對話管理系統之改善 43 4.4 各子系統之整合 44 第五章 實驗結果 47 5.1 實驗環境 47 5.2 關鍵詞數與語音辨識正確率的實驗 47 5.3 SGDS測試及錯誤回復實驗 48 5.4 MIS對話完成度實驗 49 第六章 結論及未來展望 53 6.1 結論 53 6.2 未來展望 53 參考文獻： 55 附錄 57