| DC 欄位 |
值 |
語言 |
| DC.contributor | 電機工程學系 | zh_TW |
| DC.creator | 陳子和 | zh_TW |
| DC.creator | Zi-He Chen | en_US |
| dc.date.accessioned | 2007-6-20T07:39:07Z | |
| dc.date.available | 2007-6-20T07:39:07Z | |
| dc.date.issued | 2007 | |
| dc.identifier.uri | http://ir.lib.ncu.edu.tw:444/thesis/view_etd.asp?URN=90541013 | |
| dc.contributor.department | 電機工程學系 | zh_TW |
| DC.description | 國立中央大學 | zh_TW |
| DC.description | National Central University | en_US |
| dc.description.abstract | 在公共電話網路中，語者辨認系統通常會遇到話筒不匹配和辨認語料不足的問題。為增進語者辨認系統之強健性，我們提出一融合下層聲學與上層韻律訊息之架構，利用韻律訊息特徵分析(latent prosody analysis, LPA)，量測不同語者間的韻律模型距離，並融合聲學模型(GMM)與韻律模型分數得到最後的辨識結果。LPA 主要是利用資訊檢索的概念將SID 問題轉化成全文檢索的問題，經由下列三步驟(1) 韻律訊息標示化( tokenization), (2) 韻律訊息分析(LPA)及(3)語者檢索(speaker retrieval) 實現利用韻律訊息之強健性語者辨識。
實驗使用 Handset TIMIT（HTIMIT）語料庫，以leave-one-out方式輪流使用九種不同的話筒當作未知話筒，驗證所提出之方法。實驗結果顯示，若以傳統 maximum likelihood a priori handset knowledge interpolation (ML-AKI) 的方法當作基礎(baseline)，語者辨識率將可傳統pitch-GMM或 prosody bi-gram modeling 方法優異，無論對已知話筒和未知話筒皆能有效改善系統之強健性。 | zh_TW |
| dc.description.abstract | Handsets that are not seen in the training phase (unseen handsets) are significant sources of performance degradation for speaker identification (SID) applications in the telecommunication environment. In this thesis, a novel latent prosody analysis (LPA) approach to automatically extract the most discriminative prosody cues for assisting in conventional spectral feature-based SID is proposed. The concept of the LPA approach is to transform the SID problem into a full-text document retrieval-like task via (1) prosodic contour tokenization, (2) latent prosody analysis, and (3) speaker retrieval. Experimental results of the phonetically balanced, read-speech, handset-TIMIT (HTIMIT) database demonstrated that the proposed method of fusing the LPA prosodic feature-based SID systems with maximum likelihood a priori handset knowledge interpolation (ML-AKI) spectral feature-based SID outperformed both the pitch and energy Gaussian mixture model (Pitch-GMM) and the bi-gram of the prosodic state (bi-gram) counterparts for both cases of counting all and only unseen handsets. | en_US |
| DC.subject | 語者辨識 | zh_TW |
| DC.subject | 韻律訊息 | zh_TW |
| DC.subject | speaker identification | en_US |
| DC.subject | prosodic information | en_US |
| DC.title | 利用韻律訊息之強健性語者辨識 | zh_TW |
| dc.language.iso | zh-TW | zh-TW |
| DC.title | Latent Prosody Analysis for Robust Speaker Identification | en_US |
| DC.type | 博碩士論文 | zh_TW |
| DC.type | thesis | en_US |
| DC.publisher | National Central University | en_US |