Joint Dictionary Learning-Based Non-Negative Matrix Factorization for Voice Conversion to Improve Speech Intelligibility After Oral Surgery

被引:32
作者
Fu, Szu-Wei [1 ,2 ]
Li, Pei-Chun [3 ]
Lai, Ying-Hui [4 ]
Yang, Cheng-Chien [3 ]
Hsieh, Li-Chun [3 ]
Tsao, Yu [2 ]
机构
[1] Natl Taiwan Univ, Dept Comp Sci & Informat Engn, Taipei, Taiwan
[2] Acad Sinica, Res Ctr Informat Technol Innovat, Taipei 11574, Taiwan
[3] Mackay Med Coll, Dept Audiol & Speech Language Pathol, New Taipei, Taiwan
[4] Yuan Ze Univ, Dept Elect Engn, Taoyuan, Taiwan
关键词
Joint dictionary learning; non-negative matrix factorization (NMF); sparse representation; voice conversion (VC);
D O I
10.1109/TBME.2016.2644258
中图分类号
R318 [生物医学工程];
学科分类号
0831 ;
摘要
Objective: This paper focuses on machine learning based voice conversion (VC) techniques for improving the speech intelligibility of surgical patients who have had parts of their articulators removed. Because of the removal of parts of the articulator, a patient's speech may be distorted and difficult to understand. To overcome this problem, VC methods can be applied to convert the distorted speech such that it is clear and more intelligible. To design an effective VC method, two key points must be considered: 1) the amount of training data may be limited (because speaking for a long time is usually difficult for postoperative patients); 2) rapid conversion is desirable (for better communication). Methods: We propose a novel joint dictionary learning based non-negative matrix factorization (JD-NMF) algorithm. Compared to conventional VC techniques, JD-NMF can perform VC efficiently and effectively with only a small amount of training data. Results: The experimental results demonstrate that the proposed JD-NMF method not only achieves notably higher short-time objective intelligibility (STOI) scores (a standardized objective intelligibility evaluation metric) than those obtained using the original unconverted speech but is also significantly more efficient and effective than a conventional exemplar-based NMF VC method. Conclusion: The proposed JD-NMF method may outperform the state-of-the-art exemplar-based NMF VC method in terms of STOI scores under the desired scenario. Significance: We confirmed the advantages of the proposed joint training criterion for the NMF-based VC. Moreover, we verified that the proposed JD-NMF can effectively improve the speech intelligibility scores of oral surgery patients.
引用
收藏
页码:2584 / 2594
页数:11
相关论文
共 47 条
[1]  
A. N. S. Institute, 1997, AM NAT STAND METH CA
[2]  
Aihara R., 2012, Proceedings of The 2012 Asia Pacific Signal and Information Processing Association Annual Summit and Conference, P1
[3]  
Aihara R, 2013, INT CONF ACOUST SPEE, P8037, DOI 10.1109/ICASSP.2013.6639230
[4]  
Cichocki A, 2006, LECT NOTES COMPUT SC, V3889, P32
[5]   Spectral Mapping Using Artificial Neural Networks for Voice Conversion [J].
Desai, Srinivas ;
Black, Alan W. ;
Yegnanarayana, B. ;
Prahallad, Kishore .
IEEE TRANSACTIONS ON AUDIO SPEECH AND LANGUAGE PROCESSING, 2010, 18 (05) :954-964
[6]  
Dong MH, 2015, ASIAPAC SIGN INFO PR, P488, DOI 10.1109/APSIPA.2015.7415320
[7]   Nonnegative Matrix Factorization with the Itakura-Saito Divergence: With Application to Music Analysis [J].
Fevotte, Cedric ;
Bertin, Nancy ;
Durrieu, Jean-Louis .
NEURAL COMPUTATION, 2009, 21 (03) :793-830
[8]  
Flanagan J. L., 2013, Speech analysis synthesis and perception, V3
[9]   Exemplar-Based Sparse Representations for Noise Robust Automatic Speech Recognition [J].
Gemmeke, Jort F. ;
Virtanen, Tuomas ;
Hurmalainen, Antti .
IEEE TRANSACTIONS ON AUDIO SPEECH AND LANGUAGE PROCESSING, 2011, 19 (07) :2067-2080
[10]  
Hao-Teng Fan, 2014, 2014 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), P4483, DOI 10.1109/ICASSP.2014.6854450