台語關鍵詞辨識之實作與比較

Implementation and Comparison of Keyword Spotting for Taiwanese

本論文主要探討結合語音評分與音高走勢分類器辨識，來實做台語關鍵詞辨識系統，以改進其辨識率。第一階段先分別利用了不同方法來實做台語關鍵詞辨識系統，第二階段使用語音評分與音高走勢分類器進行驗證以改善辨識效能。首先，在第一階段使用了隱藏式馬可夫模型（hidden Markov model），以及音素匹配法（phone mismatch）。而第二階段則是將此兩種方法辨識出來的候選關鍵詞，進行語音評分和音高走勢分類器驗證，將此兩種方法分別設立門檻值，利用決策樹法進行驗證。實驗結果顯示，在兩種基礎方法後，加入語音評分做為驗證的相等錯誤率（equal errorrate, EER），分別約可下降20%和5%；進一步加入音高走勢分類器驗證後，約可再下降1%，因此語音評分和音高走勢分類器對於台語關鍵詞系統的驗證是很有幫助的。

This paper focuses on improving in the performance of a Taiwanese keyword spotting system by integrating speech assessment and pitch contour classification. In the first part of this research, we use different methods to implement a Taiwanese keyword spotting system. In second part, we improve the system by validation using speech assessment and pitch contour classification. Two methods are adopted in the first part to implement the keyword spotting system: hidden Markov model and phone mismatching method. We then perform speech assessment and pitch contour classification to validate the candidate keywords selected by these two methods to refine the results. A threshold is used for a decision tree to make the final decision. Experimental results shows that the equal error rates (ERRs) reduce about 20% and 5% after being incorporated speech assessment validation. After being incorporated with pitch contour classification, ERRs further reduce about 1%. This concludes that the validation technique using speech assessment and pitch contour classification can improve the performance of Taiwanese keyword spotting.