基於隱藏式條件隨機域聲學模型之資源受限裝置語音命令系統

Mixed-Lingual Acoustic Modeling of Hidden Conditional Random Field for Resource-constrained Voice Command System

許順翔(Shun-Siang Syu)

指導教授 ： 洪維廷

元智大學/電機通訊學院/通訊工程學系/碩士(2009年)

https://doi.org/10.6838/YZU.2009.00304

摘要

本論文目的為發展適用於資源受限運算環境，結合華語、英語的混合語言關鍵詞辨識，對於隱藏式條件隨機域華語英語語音模型的差異造成條件相似度的差距，提出權重偏差補償法，並利用光束搜尋法降低搜尋辨識時所佔的記憶體空間，並維持一定的辨識率。
本論文的主要貢獻為，一、深入探討適用於隱藏式條件隨機域之定點演算法，二、提出整合搜尋偏差補償之定點化演算法，三、提出兩階段式權重偏差補償法。實驗證明本論文之方法在不同光束寬度資源受限下，與傳統HMM方法比較能顯出其強健能力的優勢。

關鍵字

隱藏式條件隨機域 ； 資源受限 ； 語音命令系統

並列摘要

The thesis presents the implementation techniques for Mandarin/English mixed-lingual speech recognition kernel under resource-constrained platforms. Considering HCRF-based conditional likelihood-gap between Mandarin and English languages, a two-stage compensation procedure is proposed to solve this bias issue in beam search. Among the related issues and techniques we explore are: (1) To derive the fixed-pointing approach for HCRF-based speech recognition. (2) Integrating the search offset compensation in fixed-pointing algorithm. (3) To explore the needed memory consumption issues in beam search between HMM-based and HCRF-based speech recognition.

並列關鍵字

Mixed-Lingual ； Hidden Conditional Random ； Resource-constrained

參考文獻

[14] 遊山銳, 簡世傑等人,”中英文混雜關鍵詞萃取技術,＂TEPS 2004, 66-79.

[1] P. C. Woodland and D. Povery, ”Large scale discriminative training of hidden Markov models for speech recognition” CSL 2002, vol. 16, 25-47, 2002.

[2] S.-B. Wang, A. Quattoni, L.P. Morency, D. Demirdjian, and T. Darrel, “Hidden condition random fields for gesture recognition”, CVPR 2006, vol. 1, 14-19, 2006.

[3] M. Mahajan, A. Gunawardana, and A. Acero, “Training algorithms for hidden conditional random fields,” ICASSP 2006, vol. 1, 14-19, 2006.

[4] Y.-H Sung, C. Boulis, C. Manning, and D. Jurafsky,”Regularization, adaptation, and non-independent features improve hidden condition random field for phone classification,” ASRU 2007, 347-352, 2007.