中文自發性語音之聲學模式及韻律模式的改進

Improved Acoustic Modeling and Prosody Modeling for Mandarin Spontaneous-Speech Recognition

游俊龍(Chung-Long You)

指導教授 ： 王逸如

國立交通大學/電機學院/電信工程系所/碩士(2015年)

https://doi.org/10.6842/NCTU.2015.00714

摘要

自發性語音(Spontaneous speech)是最接近人們日常生活的對話，因此也顯得非常重要。本研究主要分為兩個部分，分別對於自發性語音之聲學模型(Acoustic Model, AM)與韻律模型(Prosodic Model, PM)進行改善。在聲學模型方面，本研究利用朗讀式語音(Read speech)來協助訓練，我們使用模型調適(Model Adaptation)的方法，將朗讀式語音的聲學模型調適至自發性語音的聲學模型，再進一步利用Skip state HMM來改善刪除型錯誤過多的情形。
而在自發性語音韻律模型方面，本研究沿用過去所提出的階層式韻律模型(Hierarchical Prosodic Model, HPM)為基礎，來建構適合自發性語音的韻律模式，本研究對於音節韻律模型(Syllable prosodic model)進行修改，考慮其他可能的影響因子(Affecting Factor, AF)，最後，語料庫經過自動標記後，探討自發性語音中特有現象的韻律變化，並期望這些發現以及改善可幫助未來進行自發性語音相關的研究。

關鍵字

自發性語音 ； 聲學模型 ； 模型調適 ； 階層式韻律模型

並列摘要

The spontaneous speech is the closest talking way to people’s daily life, therefore it appears to be very important. This thesis has two parts, one is about improving acoustic model and the other is about improving prosody model. In acoustic modeling, we use the read speeh data to assist the training and use the method of model adaptation to adapting acoustic model of read speech to spontaneous speech. Furthermore, we use the technology of the skip state HMM to fix deletion error problem.
In prosodic modeling, we construct prododic model which is adapted to spontaneous speech based on the Hierarchical Prosodic Model (HPM). We modify syllable prosodic model and consider other possible affecting factors. Lastly, an analysis of disfluencies related to the labeling results is also discussed and we expect those results would be able to improve the research on spontaneous speech.

並列關鍵字

spontaneous speech ； acoustic model ； model adaptation ； Hierarchical Prosodic Model

參考文獻

[10] 周裕倫，“中文自發性語音之韻律標記及韻律模式”，國立交通大學碩士論文，民國九十八年七月。

[1] M. Riley, W. Byrne, M. Finke, S. Khudanpur, and A. Ljolje, “Stochastic pronunciation modeling from hand-labelled phonetic corpora,” Speech Communication, Vol. 29, No. 2-4, pp. 209-224, 1999

[2] Y. Liu, and P. Fung, “State-Dependent Phonetic Tied Mixtures with Pronunciation Modeling for Spontaneous Speech Recognition,” IEEE Transactions on Speech and Audio Processing, Vol. 12, No.4, pp. 351-364, 2004

[3] Nanjo, H.; Kawahara, T., "Language model and speaking rate adaptation for spontaneous presentation speech recognition," Speech and Audio Processing, IEEE Transactions on , vol.12, no.4, pp. 391- 400, July 2004

[9] 江振宇，“非監督式中文語音韻律標記及韻律模式”，國立交通大學博士論文，民國九十八年三月。