基於深度學習的語音人名辨識系統;Deep Learning Based Speech Personal Name Recognition System

NCU Institutional Repository > 資訊電機學院 > 資訊工程學系碩士在職專班 > 博碩士論文 > Item 987654321/89744

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/89744

题名:	基於深度學習的語音人名辨識系統;Deep Learning Based Speech Personal Name Recognition System
作者:	鄭雅馨;Cheng, Ya-Hsin
贡献者:	資訊工程學系在職專班
关键词:	語音人名辨識;自動語音辨識;命名實體識別;雙數組Trie樹;AC自動機算法
日期:	2022-07-04
上传时间:	2022-10-04 11:58:18 (UTC+8)
出版者:	國立中央大學
摘要:	公司總機人員經常需要轉接客戶來電給公司同仁，不僅耗時而且容易失誤。本研究透過命名實體識別技術，自動擷取語音中的人名，透過雙數組Trie樹與AC自動機算法的技術，融合編輯距離的方法，進而找出公司同仁的人名。我們以精確率、召回率及F1 Score為評估方法，對不同類型語料的不同語音來源，進行語音人名辨識評估。最後我們設計了一個語音人名辨識系統，模擬一間小公司的電話轉接功能，以此驗證辨識性能。實驗結果顯示，可辨認出的在職員工的準確率為90.2%，而辨識出無此員工的準確率為88.32%，而整體的準確率達到89.73%。本研究成果可應用於公司總機的自動轉機。關鍵詞：語音人名辨識；自動語音辨識、命名實體識別、雙數組Trie樹、AC自動機算法 ;The company′s switchboard often needs to transfer customer calls to the company′s colleagues, which is not only time-consuming but also prone to errors. In this study, named entity recognition technology is used to automatically capture the names of people in speech, and the Double-Array Trie and Aho–Corasick algorithm are combined with the edit distance method to find out the names of colleagues in the company. We use precision, recall and F1 Score as evaluation methods to evaluate speech name recognition for different speech sources of different types of corpus. Finally, we designed a speech personal name recognition system to simulate the phone transfer function of a small company to verify the recognition performance. The experimental results show that the accuracy of identifying active employees is 90.2%, and the accuracy of identifying ex-employees or un-hired employees is 88.32%, and the overall accuracy is 89.73%. The research results can be applied to the automatic transfer of the company′s switchboard. Keywords: speech personal name recognition; automatic speech recognition, named entity recognition, Double-Array Trie, Aho–Corasick algorithm.
显示于类别:	[資訊工程學系碩士在職專班 ] 博碩士論文

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	112	检视/开启

在NCUIR中所有的数据项都受到原著作权保护.

社群 sharing

数据加载中.....