基于自適應(yīng)逼近殘差的稀疏表示語音降噪方法
doi: 10.11999/JEIT160369
基金項(xiàng)目:
國家自然科學(xué)基金(61571192)，廣東省公益項(xiàng)目(2015A010103003)
Adapted Stopping Residue Error Based Sparse Representation for Speech Denoising
Funds:
The National Natural Science Foundation of China (61571192), The Science and Technology Foundation of Guangdong Province (2015A010103003)
-
摘要: 該文提出一種基于自適應(yīng)逼近殘差的稀疏表示語音降噪方法。在字典學(xué)習(xí)階段基于K奇異值分解(K-Singular Value Decomposition, K-SVD)算法獲得干凈語音譜的過完備字典，在稀疏表示階段基于權(quán)重因子調(diào)整后的噪聲譜和估計(jì)的交叉項(xiàng)對(duì)逼近殘差持續(xù)自適應(yīng)地更新，并采用正交匹配追蹤(Orthogonal Matching Pursuit, OMP)方法對(duì)干凈語音譜進(jìn)行稀疏重構(gòu)。最后結(jié)合估計(jì)的干凈語音譜與帶噪語音相位，通過傅里葉逆變換獲得重構(gòu)的干凈語音。實(shí)驗(yàn)結(jié)果表明所提方法在不同噪聲和信噪比條件下相比標(biāo)準(zhǔn)的譜減法，稀疏表示語音降噪算法和基于自回歸隱馬爾可夫模型的降噪方法有更好的降噪效果。Abstract: A sparse representation speech denoising method based on adapted stopping residue error is proposed. Firstly, an over complete dictionary of the clean speech power spectrum is learned by the K-Singular Value Decomposition (K-SVD) algorithm. In the sparse representation stage, the stopping residue error is adaptively achieved according to the estimated cross terms and the noise spectrum which is adjusted by a weighted factor, and the Orthogonal Matching Pursuit (OMP) approach is applied to reconstruct the clean speech spectrum from the noisy speech. Finally, the clean speech is re-synthesis via the inverse Fourier transform with the reconstructed speech spectrum and the noisy speech phase. The experiment results show that the proposed method outperforms the standard spectral subtraction, sparse representation based speech denoising algorithm and the AutoRegressive Hidden Markov Model (AR-HMM) based speech denoising method in terms of subjective and objective measure.
-
BABY D, VIRTANEN T, GEMMEKE J F, et al. Coupled dictionaries for exemplar-based speech enhancement and automatic speech recognition[J]. IEEE Transactions on Audio, Speech, and Language Processing, 2015, 23(11): 1788-1799. doi: 10.1109/TASLP.2015.2450491. ZHOU W L and HE Q H. Non-intrusive speech quality objective evaluation in high-noise environments[C]. IEEE China Summit and International Conference on Signal and Information Processing, Chengdu, China, 2015: 50-54. doi: 10.1109/ChinaSIP.2015.7230360. KODRASI I, MARQUARDT D, and DOCLO S. Curvature-based optimization of the trade-off parameter in the speech distortion weighted multichannel wiener filter[C]. IEEE International Conference on Acoustics, Speech and Signal Processing, South Brisbane, Australia, 2015: 315-319. doi: 10.1109/ICASSP.2015.7177982. MARTIN R. Noise power spectral density estimation based on optimal smoothing and minimum statistics[J]. IEEE Transactions on Speech and Language Processing, 2001, 9(5): 504-512. doi: 10.1109/89.928915. GERKMANN T. MMSE-optimal enhancement of complex speech coefficients with uncertain prior knowledge of the clean speech phase[C]. IEEE International Conference on Acoustics, Speech and Signal Processing, Florence, Italy, 2014: 4478-4482. doi: 10.1109/ICASSP.2014.6854449. DAVID Y and KLEIJN W B. HMM-based gain modeling for enhancement of speech in noise[J]. IEEE Transactions on Audio, Speech, and Language Processing, 2007, 15(3): 882-892. 10.1109/TASL.2006.885256. EVANA N, MASON J, LIU W, et al. An assessment on the fundamental limitations of spectral subtraction[C]. IEEE International Conference on Acoustics, Speech and Signal Processing, Toulous, France, 2006: 145-148. doi: 10.1109/ ICASSP.2006.1659978. HILMAN F, KOJI I, and KOICHI S. Feature normalization based on non-extensive statistics for speech recognition[J]. Speech Communication, 2013, 55(5): 587-599. doi: 10.1016/ j.specom.2013.02.004. HSIEH C T, HUANG P Y, CHEN Y H, et al. Speech enhancement based on sparse representation under color noisy environment[C]. International Symposium on Intelligent Signal Processing and Communication Systems, Nusa Dua, Indonesia, 2015: 134-138. doi: 10.1109/ISPACS. 2015.7432752. 孫林慧, 楊震. 基于數(shù)據(jù)驅(qū)動(dòng)字典和稀疏表示的語音增強(qiáng)[J]. 信號(hào)處理, 2011, 27(12): 1793-1800. SUN L H and YANG Z. Speech enhancement based on datadriven dictionary and sparse representation[J]. Signal Processing, 2011, 27(12): 1793-1800. ZHAO Y P, ZHAO X H, and WANG B. A speech enhancement method employing sparse representation of power spectral density[J]. Journal of Information and Computational Science, 2013, 10(6): 1705-1714. ZHAO N, XU X, and YANG Y. Sparse representations for speech enhancement[J]. Chinese Journal of Electronics, 2011, 19(2): 268-272. SIGG C D, DIKK T, and BUHMANN J M. Speech enhancement using generative dictionary learning[J]. IEEE Transactions on Audio, Speech, and Language Processing, 2012, 20(6): 1698-1712. doi: 10.1109/TASL.2012.2187194. ZHAO Y P and WANG B. A speech enhancement method based on sparse reconstruction of power spectral density [J]. Computers Electrical Engineering, 2014, 40(4): 1705-1714. doi: 10.1016/j.compeleceng.2013.12.007. LOIZOU P C. Speech Enhancement: Theory and Practice [M]. Florida, US: CRC Press, 2013: 104-106. RANGACHARI S and LOIZOU P. A noise estimation algorithm for highly nonstationary environments[J]. Speech Communication, 2006, 48(2): 220-231. doi: 10.1016/ j.specom.2006.08.005. BEROUTI M, SCHWARTZ M, and MAKHOUL J. Enhancement of speech corrupted by acoustic noise[C]. IEEE International Conference on Acoustics, Speech and Signal Processing, Washington, US, 1979: 4478-4482. doi: 10.1109/ ICASSP.1979.1170788. CHANG L H and WU J Y. An improved RIP-based performance guarantee for sparse signal recovery via orthogonal matching pursuit[J]. IEEE Transactions on Information Theory, 2014, 60(9): 5702-5715. doi: 10.1109/ TIT.2014.2338314. AHARON M and ELAD M. K-SVD: An algorithm for designing overcomplete dictionaries for sparse representation [J]. IEEE Transactions on Signal Processing, 2006, 54(11): 4311-4322. doi: 10.1109/TSP.2006. Signal 881199. ITU-T. P.862-2001. Perceptual evaluation of speech quality (PESQ): An objective method for end to end speech quality assessment of narrow-band telephone networks and speech codecs[S]. Geneva, ITU-T, 2001. -
計(jì)量
- 文章訪問數(shù): 1553
- HTML全文瀏覽量: 201
- PDF下載量: 509
- 被引次數(shù): 0
下載:
