自动发音错误检测中基于最大化F1值准则的区分性特征补偿训练算法

doi:10.3969/j.issn.0372-2112.2015.07.007

电子学报 ›› 2015, Vol. 43 ›› Issue (7): 1294-1299.DOI: 10.3969/j.issn.0372-2112.2015.07.007

自动发音错误检测中基于最大化F1值准则的区分性特征补偿训练算法

黄浩¹, 徐海华², 王羡慧¹, 吾守尔·斯拉木¹

1. 新疆大学信息科学与工程学院, 新疆乌鲁木齐 830046;
2. 南洋理工大学Temasek实验室, 新加坡 639798

收稿日期:2013-12-26 修回日期:2014-09-24 出版日期:2015-07-25
- 作者简介:
- 黄浩 男,1976 年10月出生,新疆乌鲁木齐人,副教授.1999 年、2004 年和2008年分别在上海交通大学、新疆大学、上海交通大学获工学学士、硕士和博士学位.主要从事语音信号处理、自然语言处理与多媒体交互方面的研究工作. E-mail:hwanghao@gmail.com;徐海华 男,1975 年2月出生,湖北黄冈人.1998年、2005 年和2010年分别在哈尔滨理工大学、华中科技大学、上海交通大学获工学学士、硕士和博士学位.目前为新加坡南洋理工大学Temasek 实验室研究科学家.主要从事语音识别、关键词检索等方面的研究工作. E-mail:haihuaxu@ntu.edu.sg;王羡慧 男,1980 年5月出生,新疆泽普人,副教授.2004年、2011年分别在新疆大学和西安交通大学获工学学士和博士学位.主要从事人工智能与机器学习等方面的研究工作. E-mail:wisdom@xju.edu.cn;吾守尔·斯拉木 男,中国工程院院士,新疆大学信息科学与工程学院教授.主要研究方向为语音识别、语音合成、多语种信息处理. E-mail:wushour@xju.edu.cn
- 基金资助:
- 国家自然科学基金 (No.61365005,No.60965002)

Maximum F1-Score Criterion Based Discriminative Feature Compensation Training Algorithm for Automatic Mispronunciation Detection

HUANG Hao¹, XU Hai-hua², WANG Xian-hui¹, Wushour Silamu¹

1. School of Information Science and Engineering, Xinjiang University, Urumqi, Xinjiang 830046, China;
2. Temasek Laboratories, Nanyang Technological University, Singapore 639798

Received:2013-12-26 Revised:2014-09-24 Online:2015-07-25 Published:2015-07-25
- Supported by:
- National Natural Science Foundation of China (No.61365005, No.60965002)

摘要/Abstract

摘要：

为提高自动发音错误检测性能,提出一种区分性特征补偿训练算法.该方法将高斯后验概率矢量经过线性变换后作为偏移量补偿至传统的谱特征.将经过正确度标注的语音数据库上的发音错误检测F1值的最大化作为变换参数的训练准则.推导了目标函数对变换参数的偏导数公式,并利用无约束参数优化例程L-BFGS更新变换参数.发音错误检测实验表明该方法能够有效增大训练和测试集的F1值.并且训练和测试集的精确度、召回率也都有明显提高.在特征优化的基础上进行模型参数训练,检错性能较单独的区分性特征训练、单独的区分性模型训练都有进一步改进.

关键词: 自动发音错误检测, F1值, 区分性训练, 特征, 计算机辅助语言学习

Abstract:

To improve the performance of automatic mispronunciation detection,a discriminative feature compensation training algorithm is proposed.The method is to train a matrix projecting from posteriors of Gaussians to a normal size feature space,and then to add the projected features to traditional spectral features.The matrix is trained according to maximum F1-score criterion,which aims at maximizing the empirical mispronunciation detection F1-score on the annotated speech database.Mispronunciation detection experiments have shown the method is effective in increasing F1-score,precision and recall on both the training data and evaluation data.It is also shown model parameter discriminative training on new features obtained further improvements over both model training and feature training.

Key words: automatic mispronunciation detection, F1-score, discriminative training, feature, computer-assisted language learning

中图分类号:

TN912.34

. 自动发音错误检测中基于最大化F1值准则的区分性特征补偿训练算法[J]. 电子学报, 2015, 43(7): 1294-1299.

. Maximum F1-Score Criterion Based Discriminative Feature Compensation Training Algorithm for Automatic Mispronunciation Detection[J]. Acta Electronica Sinica, 2015, 43(7): 1294-1299.

参考文献

[1] Wei S,Hu G P,Hu Y,Wang R H.A new method for mispronunciation detection using support vector machine based on pronunciation space models[J].Speech Communication,2009,51(10):896-905.
[2] Witt S M,Young S J.Phone-level pronunciation scoring and assessment for interactive language learning[J].Speech Communication,2000,30 (2-3):95-108.
[3] 葛凤培,潘复平,董滨,颜永红.汉语发音质量评估的实验研究[J].声学学报,2010,35(2):261-266. GE Feng-pei,PAN Fu-ping,DONG Bin,YAN Yong-hong.Experiment investigation of Putonghua quality assessment[J].Acta Acustica,2010,35(2):261-266.(in Chinese)
[4] Lo W K,Zhang S,Meng H.Automatic derivation of phonological rules for mispronunciation detection in a computer-assisted pronunciation training system[A].Proceedings of Interspeech[C].JAPAN:ISCA,2010.765-768.
[5] Qian X,Soong F,Meng H.Discriminative acoustic model for improving mispronunciation detection and diagnosis in computer-aided pronunciation training (CAPT)[A].Proceedings of Interspeech[C].JAPAN:IEEE,2010.757-760.
[6] Luo D,Yang X,Wang L.Improvement of segmental mispronunciation detection with prior knowledge extracted from large L2 speech corpus[A].Proceedings of Interspeech[C].Italy:ISCA,2011.1593-1596.
[7] Qian X,Meng H,Soong F.The use of DBN-HMMs for mispronunciation detection and diagnosis in L2 English to support computer-aided pronunciation training[A].Proceedings of Interspeech[C].USA:ISCA,2012.775-778.
[8] Lee A,Zhang Y,Glass J.Mispronunciation detection via dynamic time warping on deep belief network-based posteriorgrams[A].Proceedings of ICASSP[C].Canada:IEEE,2013.8227-8231.
[9] Hu W,Qian Y,Soong F.A new DNN-based high quality pronunciation evaluation for computer-aided language learning (CALL)[A].Proceedings of Interspeech[C].France:ISCA,2013.1886-1890.
[10] Juang B H,Katagiri S.Discriminative learning for minimum error classification[J].IEEE Transactions on Signal Processing,1992,40(12):3043-3054.
[11] Bahl L R,Brown P F,Souza P,Mercer R.Maximum mutual information estimation of hidden Markov model parameters for speech recognition[A].Proceedings of ICASSP[C].JAPAN:IEEE,1986.49-52.
[12] Povey D,Woodland P.Minimum phone error and I-smoothing for improved discriminative training[A].Proceedings of ICASSP[C].USA:IEEE,2002.105-108.
[13] Huang H,Wang J,Abudureyimu H.Maximum F1-score discriminative training for automatic mispronunciation detection in computer-assisted language learning[A].Proceedings of Interspeech[C].USA:ISCA.2012.815-818.
[14] Droppo J,Acero A.Maximum mutual information SPLICE transform for seen and unseen conditions[A].Proceedings of Interspeech[C].Portugal:ISCA,2005.989-992.
[15] Povey D,Kingsbury B,Mangu L,et al.fMPE:discriminatively trained features for speech recognition[A].Proceedings of ICASSP[C].USA:IEEE,2005.961-964.
[16] Zhang B,Matsoukas S,Schwartz R.Discriminatively trained region dependent feature transforms for speech recognition[A].Proceedings of ICASSP[C].FRANCE:IEEE,2006.I313-I316.
[17] Nocedal J,Wright S J.Numerical Optimization[M].Germany:Springer,1999.
[18] 竺博.区分性训练和区分性自适应在自动语音识别声学模型优化中的应用[D].安徽合肥:中国科学技术大学,2009. ZHU Bo.Application of discriminative training and discriminative training-adaptation in acoustic modeling of ASR[D].Hefei,Anhui:University of Science and Technology of China,2009.(in Chinese)

自动发音错误检测中基于最大化F1值准则的区分性特征补偿训练算法

Maximum F1-Score Criterion Based Discriminative Feature Compensation Training Algorithm for Automatic Mispronunciation Detection

PDF

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 15

编辑推荐

Metrics

[1]	范兵兵, 何庭建, 张聪炫, 陈震, 黎明. 联合遮挡约束与残差补偿的特征金字塔光流计算方法[J]. 电子学报, 2023, 51(3): 648-657.
[2]	王俊杰, 冯德军, 隋冉, 邢世其, 肖顺平. 基于非周期PSS的SAR目标特征操控方法研究[J]. 电子学报, 2023, 51(3): 564-572.
[3]	杨利平, 侯振威, 辜小花, 郝峻永. 弱标签声音事件检测的空间-通道特征表征与自注意池化[J]. 电子学报, 2023, 51(2): 297-306.
[4]	张聿远, 闫文君, 张立民. 基于多模态特征融合网络的空时分组码识别算法[J]. 电子学报, 2023, 51(2): 489-498.
[5]	许新征, 李杉. 基于特征膨胀卷积模块的轻量化技术研究[J]. 电子学报, 2023, 51(2): 355-364.
[6]	高继蕊, 李华锋, 张亚飞, 谢明鸿, 李凡. 双注意力引导的细节和结构信息融合图像去雾网络[J]. 电子学报, 2023, 51(1): 160-171.
[7]	李滔, 董秀成, 林宏伟. 基于深监督跨尺度注意力网络的深度图像超分辨率重建[J]. 电子学报, 2023, 51(1): 128-138.
[8]	柳长源, 张玉亮, 毕晓君. 基于多阶段提议稀疏区域卷积网络的城市交通目标检测[J]. 电子学报, 2023, 51(1): 26-31.
[9]	王硕, 王坚, 王亚男, 宋亚飞. 一种基于特征融合的恶意代码快速检测方法[J]. 电子学报, 2023, 51(1): 57-66.
[10]	琚长瑞, 秦晓燕, 袁广林, 李豪, 朱虹. 尺度敏感损失与特征融合的快速小目标检测方法[J]. 电子学报, 2022, 50(9): 2119-2126.
[11]	任莎莎, 刘琼. 小目标特征增强图像分割算法[J]. 电子学报, 2022, 50(8): 1894-1904.
[12]	王飞扬, 冀鹏欣, 孙笠, 危倩, 李根, 张忠宝. 一种基于深度学习的动态社交网络用户对齐方法[J]. 电子学报, 2022, 50(8): 1925-1936.
[13]	邵志文, 周勇, 谭鑫, 马利庄, 刘兵, 姚睿. 基于深度学习的表情动作单元识别综述[J]. 电子学报, 2022, 50(8): 2003-2017.
[14]	强幸子, 金翔, 张天骐. 基于相似度的NPLC-DSSS信号扩频码盲估计[J]. 电子学报, 2022, 50(8): 2043-2048.
[15]	江兵兵, 何文达, 吴兴宇, 项俊浩, 洪立斌, 盛伟国. 基于自适应图学习的半监督特征选择[J]. 电子学报, 2022, 50(7): 1643-1652.