An Adaptation Method in Noise Mismatch Conditions for DNN-based Speech Enhancement = An Adaptation Method in Noise Mismatch Conditions for DNN-based Speech Enhancement

다국어 초록 (Multilingual Abstract)

The deep learning based speech enhancement has shown considerable success. However, it still suffers performance degradation under mismatch conditions. In this paper, an adaptation method is proposed to improve the performance under noise mismatch conditions. Firstly, we advise a noise aware training by supplying identity vectors (i-vectors) as parallel input features to adapt deep neural network (DNN) acoustic models with the target noise. Secondly, given a small amount of adaptation data, the noise-dependent DNN is obtained by using L2 regularization from a noise-independent DNN, and forcing the estimated masks to be close to the unadapted condition. Finally, experiments were carried out on different noise and SNR conditions, and the proposed method has achieved significantly 0.1%-9.6% benefits of STOI, and provided consistent improvement in PESQ and segSNR against the baseline systems.

참고문헌 (Reference)

1 Bingyin Xia, "Wiener filtering based speech enhancement with weighted denoising auto-encoder and noise classification" 60 : 13-29, 2014

2 Zechao Li, "Weakly-supervised Deep Matrix Factorization for Social Image Understanding" 1-13, 2016

3 Zechao Li, "Weakly Supervised Deep Metric Learning for Community-Contributed Image Retrieval" 1989-1999, 2015

4 Timo Gerkmann, "Unbiased mmse-based noise power eatimation with low complexity and low tracking delay" 20 (20): 1383-1393, 2012

5 J. Stadermann, "Two-stage speaker adaptation of hybrid tied-posterior acoustic models" I : 997-1000, 2005

6 Stadermann, J., "Two-stage speaker adaptation of hybrid tied-posterior acoustic models" 2005

7 C. Veaux, "The voice bank corpus: Design, collection and data analysis of a large regional accent speech database" 1-4, 2013

8 Steven Boll, "Suppression of acoustic noise in speech using spectral subtraction" 27 (27): 113-122, 1979

9 Y. Ephraim, "Statistical-model-based speech enhancement systems" 80 (80): 1526-1555, 1992

10 Kim, D. Y., "Speech recognition in noisy environments using first-order vector Taylor Series" 24 (24): 39-49, 1998

11 Bing-yin Xia, "Speech enhancement with weighted denoising auto-encoder" INTERSPEECH 3444-3448, 2013

12 Cohen, "Speech enhancement for nonstationary noise environments" 81 (81): 2403-2418, 2001

13 Xugang Lu, "Speech enhancement based on deep denoising autoencoder" INTERSPEECH 436-440, 2013

14 Wang D, "Speech Separation by Humans and Machines" Kluwer 181-197, 2005

15 Philipos. C. Loizou, "Speech Enhancement: Theory and Practice" CRC Press, Inc 2013

16 George Saon, "Speaker adaptation of neural network acoustic models using i-vectors" 55-59, 2013

17 R. Talmon, "Single-channel transient interference suppression with diffusion maps" 21 (21): 132-144, 2013

18 Glembek, O., "Simplification and optimization of i-vector extraction" 4515-4519, 2011

19 Kjems U, "Role of mask pattern in intelligibility of ideal binary-masked noisy speech" 126 : 1415-1426, 2009

20 Kevin W Wilson, "Regularized non-negative matrix factorization with temporal dependencies for speech denoising" Interspeech 411-414, 2008

21 Li, X., "Regularized adaptation of discriminative classifiers" 1 : I-I, 2006

22 Hui Zou, "Regularization and variable selection via the elastic net" 67 (67): 301-320, 2005

23 Tibshirani, R., "Regression shrinkage and selection via the lasso" 58 (58): 267-288, 1996

24 H. Hermansky, "RASTA processing of speech" 2 (2): 578-589, 1994

25 H. Hermansky, "Perceptual linear predictive(PLP)analysis of speech" 87 : 1738-1752, 1990

26 A. Rix, "Perceptual evaluation of speech quality(PESQ)-a new method for speech quality assessment of telephone networks and codecs" 749-752, 2001

27 Yuxuan Wang, "On Training Target for Supervised Speech Separation" 22 (22): 1849-1858, 2014

28 Y. Bengio, "Learning deep architectures for AI" 2 (2): 1-127, 2009

29 Yu, D., "Kl-divergence regularized deep neural network adaptation for improved large vocabulary speech recognition" 7892-7897, 2013

30 Hinton G.E, "Improving neural networks by preventing co-adaptation of feature detectors" Cornell University 2013

31 Miao Yajie, "Improving Low-Resource CD-DNN-HMM using Dropout and Multilingual DNN Training" ISCA 2237 (2237): 2237-2241, 2013

32 Li, J., "HMM adaptation using a phase-sensitive acoustic distortion model for environment-robust speech recognition" 4069-4072, 2008

33 Dehak, N., "Front-end factor analysis for speaker verification" 19 (19): 788-798, 2011

34 Yu, D., "Feature learning in deep neural networks-studied on speech recognition tasks" 2013

35 Seide, F., "Feature engineering in context-dependent deep neural networks for conversational speech transcription" 24-29, 2011

36 Abdel-Hamid, O., "Fast speaker adaptation of hybrid NN/HMM model for speech recognition based on discriminative learning of speaker code" 7942-7946, 2013

37 Li N, "Factors influencing intelligibility of ideal binary masked speech : Implications for noise reduction" 123 (123): 1673-1682, 2008

38 Christine De Mol, "Elastic-net regularization in learning theory" 25 (25): 201-230, 2009

39 Shaofei Xue, "Direct Adaptation of Hybrid DNN/HMM Model for Fast Speaker Adaptation in LVCSR Based on Speaker Code" 6389 (6389): 6389-6393, 2013

40 D. L. Wang, "Computational auditory scene analysis: Principles, algorithms, and applications" Wiley-IEEE Press 2006

41 D. K. Kim, "Baysian speaker adaptation based on probabilistic principal component analysis" INTERSPEECH 734-737, 2000

42 A.Varga, "Assessment for automatic speech recognition: II. NOISEX-92: A database and an experiment to study the effect of additive noise on speech recognition systems" 12 : 247-251, 1993

43 Seltzer, M., "An investigation of deep neural networks for noise robust speech recognition" 2013

44 G. Kim, "An algorithm that improves speech intelligibility in noise for normal-hearing listeners" 126 : 1486-1494, 2009

45 C. Taal, "An algorithm for intelligibility prediction of time-frequency weighted noisy speech" 19 (19): 2125-2136, 2011

46 J. Lim, "All-pole modeling of degraded speech" 26 (26): 197-210, 1978

47 J. Duchi, "Adaptive subgradient methods for online learning and stochastic optimization" 2121-2159, 2011

48 Albensano, D., "Adaptation of artificial neural networks avoiding catastrophic forgetting" 1554-1561, 2006

49 Scott Pennock, "Accuracy of the perceptual evaluation of speech quality (pesq) algorithm" 25 : 2002

50 Moreno, P. J., "A vector Taylor series approach for environment-independent speech recognition" 733-736, 1996

51 Li, J., "A unified framework of HMM adaptation with joint compensation of additive and convolutive distortions via vector Taylor series" 65-70, 2007