| { |
| "paper_id": "O17-1030", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T07:59:31.496777Z" |
| }, |
| "title": "Development of a software-based User-Interface of Speech Enhancement System", |
| "authors": [ |
| { |
| "first": "Tao-Wei", |
| "middle": [], |
| "last": "\u738b\u97dc\u7dad", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "\u66f9\u6631", |
| "middle": [], |
| "last": "Wang", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Tsao", |
| "middle": [], |
| "last": "Yu", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "\u4e2d\u592e\u7814\u7a76\u9662\u8cc7\u8a0a\u5275\u65b0\u79d1\u6280\u7814\u7a76\u4e2d\u5fc3", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Ying-Hui", |
| "middle": [], |
| "last": "\u8cf4\u7a4e\u6689", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Lai", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "Chia-Lung", |
| "middle": [], |
| "last": "\u5433\u5bb6\u9686", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "\u8a31\u7965\u5e73 Hsiang-Ping", |
| "middle": [], |
| "last": "Wu", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "\u6cd5\u52d9\u90e8\u8abf\u67e5\u5c40", |
| "middle": [], |
| "last": "Hsu", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "The topic is to develop a user interface of speech enhancement in this study. This system includes of typical and based-on machine learning algorithm and provides a convenient and user-friendly interface. User can obtain waveform and spectrogram of enhancement speech 323", |
| "pdf_parse": { |
| "paper_id": "O17-1030", |
| "_pdf_hash": "", |
| "abstract": [ |
| { |
| "text": "The topic is to develop a user interface of speech enhancement in this study. This system includes of typical and based-on machine learning algorithm and provides a convenient and user-friendly interface. User can obtain waveform and spectrogram of enhancement speech 323", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Abstract", |
| "sec_num": null |
| } |
| ], |
| "body_text": [ |
| { |
| "text": "[2] Scalart, P., et al., \"Speech enhancement based on a priori signal to noise estimation,\" in Proc. ICASSP, pp. 629-632, 1996.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "Modulation Wiener filter for improving speech intelligibility", |
| "authors": [ |
| { |
| "first": "C.-C", |
| "middle": [], |
| "last": "Hsu", |
| "suffix": "" |
| }, |
| { |
| "first": "K.-M", |
| "middle": [], |
| "last": "Cheong", |
| "suffix": "" |
| }, |
| { |
| "first": "J.-T", |
| "middle": [], |
| "last": "Chien", |
| "suffix": "" |
| }, |
| { |
| "first": "Chi", |
| "middle": [], |
| "last": "", |
| "suffix": "" |
| }, |
| { |
| "first": "T.-S", |
| "middle": [], |
| "last": "", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)", |
| "volume": "", |
| "issue": "", |
| "pages": "370--374", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Hsu, C.-C., Cheong, K.-M., Chien, J.-T., and Chi, T.-S, \"Modulation Wiener filter for improving speech intelligibility,\" IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 370-374, 2015", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "Suppression of acoustic noise in speech using spectral subtraction", |
| "authors": [ |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Boll", |
| "suffix": "" |
| } |
| ], |
| "year": 1979, |
| "venue": "IEEE Transactions on Acoustics, Speech and Signal Processing", |
| "volume": "27", |
| "issue": "2", |
| "pages": "113--120", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Boll, S., \"Suppression of acoustic noise in speech using spectral subtraction,\" IEEE Transactions on Acoustics, Speech and Signal Processing, vol. 27, no. 2, pp. 113-120, 1979.", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "Speech enhancement using a minimum-mean square error short-time spectral amplitude estimator", |
| "authors": [ |
| { |
| "first": "Y", |
| "middle": [], |
| "last": "Ephraim", |
| "suffix": "" |
| }, |
| { |
| "first": "D", |
| "middle": [], |
| "last": "Malah", |
| "suffix": "" |
| } |
| ], |
| "year": 1984, |
| "venue": "IEEE Transactions on Audio, Speech, and Language Processing", |
| "volume": "32", |
| "issue": "6", |
| "pages": "1109--1121", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Ephraim, Y. and Malah, D., \"Speech enhancement using a minimum-mean square error short-time spectral amplitude estimator,\" IEEE Transactions on Audio, Speech, and Language Processing, vol. 32, no. 6, pp. 1109-1121, 1984.", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "Generalized Maximum a Posteriori Spectral Amplitude Estimation for Speech Enhancement", |
| "authors": [ |
| { |
| "first": "Y", |
| "middle": [], |
| "last": "Tsao", |
| "suffix": "" |
| }, |
| { |
| "first": "Y.-H", |
| "middle": [], |
| "last": "Lai", |
| "suffix": "" |
| } |
| ], |
| "year": 2016, |
| "venue": "Speech Communication", |
| "volume": "76", |
| "issue": "", |
| "pages": "112--126", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Tsao, Y. and Lai, Y.-H., \"Generalized Maximum a Posteriori Spectral Amplitude Estimation for Speech Enhancement,\" Speech Communication, vol. 76, pp. 112-126, 2016.", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "Speech enhancement using a soft decision noise suppression filter", |
| "authors": [ |
| { |
| "first": "R", |
| "middle": [], |
| "last": "Mcaulay", |
| "suffix": "" |
| }, |
| { |
| "first": "M", |
| "middle": [], |
| "last": "Malpass", |
| "suffix": "" |
| } |
| ], |
| "year": 1980, |
| "venue": "IEEE Transactions on Acoustics, Speech and Signal Processing", |
| "volume": "28", |
| "issue": "2", |
| "pages": "137--145", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "McAulay, R. and Malpass, M., \"Speech enhancement using a soft decision noise suppression filter,\" IEEE Transactions on Acoustics, Speech and Signal Processing, vol. 28, no. 2, pp. 137-145, 1980.", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "Speech enhancement by map spectral amplitude estimation using a super-Gaussian speech model", |
| "authors": [ |
| { |
| "first": "T", |
| "middle": [], |
| "last": "Lotter", |
| "suffix": "" |
| }, |
| { |
| "first": "P", |
| "middle": [], |
| "last": "Vary", |
| "suffix": "" |
| } |
| ], |
| "year": 2005, |
| "venue": "EURASIP journal on applied signal processing", |
| "volume": "", |
| "issue": "", |
| "pages": "1110--1126", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Lotter, T. and Vary, P., \"Speech enhancement by map spectral amplitude estimation using a super-Gaussian speech model,\" EURASIP journal on applied signal processing, vol. 2005, pp. 1110-1126, 2005.", |
| "links": null |
| }, |
| "BIBREF6": { |
| "ref_id": "b6", |
| "title": "Multichannel nonnegative matrix factorization in convolutive mixtures for audio source separation", |
| "authors": [ |
| { |
| "first": "A", |
| "middle": [], |
| "last": "Ozerov", |
| "suffix": "" |
| }, |
| { |
| "first": "C", |
| "middle": [], |
| "last": "F\u00b4evotte", |
| "suffix": "" |
| } |
| ], |
| "year": 2010, |
| "venue": "IEEE Transactions on Audio, Speech, and Language Processing", |
| "volume": "18", |
| "issue": "3", |
| "pages": "550--563", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Ozerov, A. and F\u00b4evotte, C., \"Multichannel nonnegative matrix factorization in convolutive mixtures for audio source separation,\" IEEE Transactions on Audio, Speech, and Language Processing, vol. 18, no. 3, pp. 550-563, 2010.", |
| "links": null |
| }, |
| "BIBREF7": { |
| "ref_id": "b7", |
| "title": "Layered nonnegative matrix factorization for speech separation", |
| "authors": [ |
| { |
| "first": "C.-C", |
| "middle": [], |
| "last": "Hsu", |
| "suffix": "" |
| }, |
| { |
| "first": "J.-T", |
| "middle": [], |
| "last": "Chien", |
| "suffix": "" |
| }, |
| { |
| "first": "Chi", |
| "middle": [], |
| "last": "", |
| "suffix": "" |
| }, |
| { |
| "first": "T.-S", |
| "middle": [], |
| "last": "", |
| "suffix": "" |
| } |
| ], |
| "year": 2015, |
| "venue": "InterSpeech", |
| "volume": "", |
| "issue": "", |
| "pages": "628--632", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Hsu, C.-C., Chien, J.-T., and Chi, T.-S., \"Layered nonnegative matrix factorization for speech separation,\" InterSpeech, pp. 628-632, 2015", |
| "links": null |
| }, |
| "BIBREF8": { |
| "ref_id": "b8", |
| "title": "Ensemble modeling of denoising autoencoder for speech spectrum restoration", |
| "authors": [ |
| { |
| "first": "X", |
| "middle": [], |
| "last": "Lu", |
| "suffix": "" |
| }, |
| { |
| "first": "Y", |
| "middle": [], |
| "last": "Tsao", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Matsuda", |
| "suffix": "" |
| }, |
| { |
| "first": "C", |
| "middle": [], |
| "last": "Hori", |
| "suffix": "" |
| } |
| ], |
| "year": 2014, |
| "venue": "Proc. INTERSPEECH", |
| "volume": "", |
| "issue": "", |
| "pages": "885--889", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Lu, X., Tsao, Y., Matsuda, S., and Hori, C., \"Ensemble modeling of denoising autoencoder for speech spectrum restoration,\" in Proc. INTERSPEECH, pp. 885-889, 2014.", |
| "links": null |
| }, |
| "BIBREF9": { |
| "ref_id": "b9", |
| "title": "Fully complex deep neural network for phase-incorporating monaural source separation", |
| "authors": [ |
| { |
| "first": "Y.-S", |
| "middle": [], |
| "last": "Lee", |
| "suffix": "" |
| }, |
| { |
| "first": "C.-Y", |
| "middle": [], |
| "last": "Wang", |
| "suffix": "" |
| }, |
| { |
| "first": "S.-F", |
| "middle": [], |
| "last": "Wang", |
| "suffix": "" |
| }, |
| { |
| "first": "J.-C", |
| "middle": [], |
| "last": "Wang", |
| "suffix": "" |
| }, |
| { |
| "first": "Wu C.-H", |
| "middle": [], |
| "last": "", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "Proceedings of ICASSP2017", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Lee, Y.-S., Wang, C.-Y., Wang, S.-F., Wang, J.-C., and Wu C.-H., \"Fully complex deep neural network for phase-incorporating monaural source separation,\" Proceedings of ICASSP2017, New Orleans, USA, March 5~9, 2017", |
| "links": null |
| }, |
| "BIBREF10": { |
| "ref_id": "b10", |
| "title": "Mood detection from daily conversational speech using denoising autoencoder and LSTM", |
| "authors": [ |
| { |
| "first": "K.-Y", |
| "middle": [], |
| "last": "Huang", |
| "suffix": "" |
| }, |
| { |
| "first": "C.-H", |
| "middle": [], |
| "last": "Wu", |
| "suffix": "" |
| }, |
| { |
| "first": "M.-H", |
| "middle": [], |
| "last": "Su", |
| "suffix": "" |
| }, |
| { |
| "first": "H.-C", |
| "middle": [], |
| "last": "Fu", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "Proceedings of ICASSP2017", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Huang, K.-Y., Wu, C.-H., Su, M.-H., and Fu, H.-C, \"Mood detection from daily conversational speech using denoising autoencoder and LSTM,\" in Proceedings of ICASSP2017, New Orleans, USA, March 5~9, 2017", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "FIGREF0": { |
| "text": "Su, Y.-C., Tsao, Y., Wu, J.-E., Jean, F.-R., \"Speech enhancement using generalized maximum a posteriori spectral amplitude estimator,\" in Proc. ICASSP, pp. 7467-7471, 2013.", |
| "type_str": "figure", |
| "num": null, |
| "uris": null |
| }, |
| "TABREF0": { |
| "num": null, |
| "text": "\u57fa\u5e95\u77e9\u9663 W \u7684\u7dad\u5ea6\u70ba F\u00d7R\u3001\u7de8\u78bc\u77e9\u9663 H \u7684\u7dad\u5ea6\u70ba R\u00d7T \u8207\u8f38\u5165\u983b\u8b5c V \u7684\u7dad\u5ea6\u70ba F\u00d7T\uff1b\u4e14 \u7df4\u8a9e\u6599\uff0c\u53d6\u5f97\u4ee3\u8868\u8a9e\u97f3\u8207\u96dc\u8a0a\u7684\u57fa\u5e95\u77e9\u9663 \u8207 \u3002\u5728\u6e2c\u8a66\u968e\u6bb5\uff0c\u5e36\u566a\u8072\u97f3\u983b\u8b5c Y \u85c9\u7531", |
| "html": null, |
| "content": "<table><tr><td>\u826f\u9019\u4e9b\u5178\u578b\u7684\u8a9e\u97f3\u589e\u5f37\u65b9\u6cd5[3]\uff0c\u63d0\u5347\u964d\u566a\u7684\u80fd\u529b\u3002\u800c\u6700\u5927\u4e8b\u5f8c\u983b\u8b5c\u632f\u5e45\u9810\u4f30\u5668\u7684\u8a9e\u97f3 \u6642\u57df\u8a0a\u865f\u3002 \u56db\u3001\u8a9e\u97f3\u589e\u5f37\u65b9\u6cd5\u6bd4\u8f03</td></tr><tr><td>\u589e\u5f37\u6f14\u7b97\u6cd5(GMAPA)\u662f\u672c\u7814\u7a76\u5718\u968a\u904e\u53bb\u7684\u7814\u7a76\u6210\u679c\uff0c\u7d50\u5408 MLSA [7] \u548c MAPA [8] \u7531\u96dc\u8a0a\u8a9e\u97f3\u4f30\u6e2c\u4e7e\u6de8\u8a9e\u97f3\u53ef\u8996\u70ba\u4e00\u500b\u51fd\u6578\u8fd1\u4f3c\u7684\u554f\u984c\uff0c\u8a72\u51fd\u6578\u5247\u7528\u4f86\u63cf\u8ff0\u96dc\u8a0a\u8a9e\u97f3 \u70ba\u4e86\u9a57\u8b49\u6b64\u4ecb\u9762\u7684\u6b63\u78ba\u6027\uff0c\u6211\u5011\u4f7f\u7528 TMHINT \u7684\u8a9e\u6599\u70ba\uff0c\u4ee5-5dB\u30010dB\u30015dB \u7684 DDAE GMAPA</td></tr><tr><td>\u5169\u7a2e\u566a\u97f3\u9810\u4f30\u6a21\u578b\u7684\u6f14\u7b97\u6cd5\u3002MLSA \u548c MAPA \u5169\u7a2e\u566a\u97f3\u9810\u4f30\u6a21\u578b\u5404\u6709\u512a\u7f3a\u9ede\uff0cMLSA \u8207 \u53d6\u5f97\u7de8\u78bc\u77e9\u9663 H\uff0c\u5982\u5f0f(2)\u6240\u793a\u3002 \u8207\u4e7e\u6de8\u8a9e\u97f3\u4e4b\u9593\u7684\u6620\u5c04\u95dc\u4fc2\u3002\u50b3\u7d71\u589e\u5f37\u6280\u8853\u4f7f\u7528\u7dda\u6027\u6620\u5c04\u51fd\u6578\uff0c\u4f46\u7531\u65bc\u8fd1\u4f86\u6df1\u5ea6\u5b78\u7fd2\u5728 SNR \u503c\u5206\u5225\u6df7\u4e0a\u8eca\u566a\u97f3\u8207\u5b30\u5152\u54ed\u8072\u566a\u97f3\u505a\u70ba\u6e2c\u8a66\u8a9e\u6599\uff0c\u900f\u904e\u672c\u7814\u7a76\u6240\u5efa\u7acb\u7684\u8a9e\u97f3\u589e\u5f37</td></tr><tr><td>\u6a21\u578b\u5c0d\u65bc\u8a0a\u865f\u522a\u6e1b\u5e45\u5ea6\u8f03\u4f4e\uff0c\u56e0\u6b64\u5728\u8a0a\u865f\u54c1\u8cea\u8f03\u4f73\u7684\u74b0\u5883\u53ef\u4ee5\u4fdd\u7559\u8f03\u591a\u7684\u8a9e\u97f3\u8cc7\u8a0a\uff0c \u8a0a\u865f\u8655\u7406\u53ca\u7269\u9ad4\u8b58\u5225\u7684\u767c\u5c55\uff0c\u4ee5\u985e\u795e\u7d93\u7db2\u8def\u7b49\u8907\u96dc\u7684\u975e\u7dda\u6027\u6620\u5c04\u51fd\u6578\u70ba\u57fa\u790e\u7684\u964d\u566a\u6cd5\u53d7 \u4ecb\u9762\u9032\u884c\u8655\u7406\u3002\u5728 NMF \u65b9\u6cd5\u7684\u7dad\u5ea6\u8a2d\u5b9a\u70ba 300\u3001100 \u8ff4\u5708\u6578\uff0c\u8a13\u7df4\u968e\u6bb5\u9078\u7528 320 \u53e5\u4e7e</td></tr><tr><td>\u4f46\u5728\u8a0a\u865f\u54c1\u8cea\u8f03\u5dee\u7684\u72c0\u6cc1\u5c0d\u65bc\u96dc\u8a0a\u6d88\u9664\u80fd\u529b\u8f03\u4f4e\uff1b\u800c MAPA \u6a21\u578b\u5c0d\u65bc\u96dc\u8a0a\u7684\u522a\u9664\u80fd\u529b (2) \u5230\u95dc\u6ce8\u3002\u4f7f\u7528\u6df1\u5c64\u53bb\u566a\u81ea\u7de8\u78bc\u6a21\u578b\u7684\u512a\u9ede\u662f\u6613\u65bc\u64f4\u5c55\u70ba\u4e0d\u540c\u5730\u985e\u795e\u7d93\u7db2\u8def\u67b6\u69cb\u4f86\u6539\u5584\u5176 \u6de8\u8a9e\u97f3\u8207\u6b32\u8655\u7406\u7684\u566a\u97f3\u8a13\u7df4\u57fa\u5e95\u77e9\u9663\uff1bDDAE \u7684\u6a21\u578b\u67b6\u69cb\u70ba[400 200 100 200 400]\u7684\u4e94 \u8868\u4e8c \u5404\u8a9e\u97f3\u589e\u5f37\u65b9\u6cd5\u5c0d\u8eca\u566a\u97f3\u8655\u7406\u5f8c\u7684 PESQ NMF</td></tr><tr><td>\u8f03\u597d\uff0c\u4f46\u5c0d\u65bc\u9ad8\u8a0a\u96dc\u6bd4\u7684\u8a0a\u865f\u5bb9\u6613\u904e\u5ea6\u522a\u6e1b\uff0c\u9020\u6210\u8f03\u5927\u7684\u8a0a\u865f\u5931\u771f\u3002GMAPA \u6f14\u7b97\u6cd5 \u6700\u5f8c\uff0c\u589e\u5f37\u5f8c\u7684\u8072\u97f3\u983b\u8b5c\u53ef\u7531\u5f0f(3)\u6c42\u5f97\u3002 \u8868\u73fe\u3002\u4f46\u4e0d\u8ad6\u662f\u57fa\u65bc\u6df1\u5c64\u53bb\u566a\u81ea\u7de8\u78bc\u6a21\u578b\u6216\u985e\u795e\u7d93\u7db2\u8def\u6a21\u578b\u964d\u566a\u6cd5\uff0c\u5b78\u7fd2\u96dc\u8a0a\u8a9e\u97f3\u53ca\u4e7e \u5c64\u6a21\u578b\uff0c\u8a13\u7df4\u8cc7\u6599\u70ba 7dB\u30013dB\u3001-3dB\u3001-7dB \u7684\u6df7\u566a\u8a9e\u97f3-\u4e7e\u6de8\u8a9e\u97f3\u5c0d\uff0c\u8a13\u7df4\u7684\u566a\u97f3\u9078</td></tr><tr><td>\u7528\u8f03\u5927\u7684\u4e8b\u524d\u6a5f\u7387\u6bd4\u4f8b\uff0c\u4ee5\u63d0\u5347\u589e\u5f37\u5f8c\u8a9e\u97f3\u8a0a\u865f\u7684\u8a0a\u96dc\u6bd4\u3002\u6b64\u5916\uff0c\u6839\u64da\u8a9e\u97f3\u8a0a\u865f\u7684\u8a0a \u96a8\u8457\u6a5f\u5668\u5b78\u7fd2(machine learning)\u7684\u9032\u5c55\uff0c\u8a9e\u97f3\u589e\u5f37\u7684\u6548\u80fd\u5df2\u7d93\u6709\u5927\u5e45\u7684\u63d0\u6607\uff0c\u5728\u773e PESQ \u7576\u4f5c\u5ba2\u89c0\u8a55\u91cf\u6307\u6a19\uff0c\u986f\u793a\u65bc\u8868\u4e8c\u548c\u8868\u56db\u3002\u8868\u4e00\u3001\u8868\u4e8c\u70ba\u8eca\u566a\u97f3\u800c\u8868\u4e09\u3001\u8868\u56db\u70ba\u5b30 \u4e00\u3001\u7dd2\u8ad6 \u4f7f\u7528\u52d5\u614b\u8abf\u6574\u4e8b\u524d\u6a5f\u7387\u6bd4\u4f8b\u7684\u6a5f\u5236\uff0c\u5728\u8f03\u9ad8\u8a0a\u96dc\u6bd4\u7684\u689d\u4ef6\u4e0b\uff0cGMAPA \u63a1\u7528\u8f03\u5c0f\u7684\u4e8b \u524d\u6a5f\u7387\u6bd4\u4f8b\uff0c\u4ee5\u9632\u6b62\u904e\u5ea6\u8a9e\u97f3\u5931\u771f\u3002\u53e6\u4e00\u65b9\u9762\uff0c\u5728\u8f03\u4f4e\u8a0a\u96dc\u6bd4\u7684\u689d\u4ef6\u4e0b\uff0cGMAPA \u4f7f , \u4e09\u3001\u8a9e\u97f3\u589e\u5f37\u4f7f\u7528\u8005\u4ecb\u9762 \u8a66\u8a9e\u6599\u8207\u8a13\u7df4\u8a9e\u6599\u76f8\u540c\uff0c\u5176\u7d50\u679c\u4ee5\u8a9e\u97f3\u6ce2\u5f62\u5716\u548c\u8072\u8b5c\u5716(0dB)\u5448\u73fe\u65bc\u8868\u4e00\u8207\u8868\u4e09\uff0c\u7528 (3) \u6de8\u8a9e\u97f3\u4e4b\u9593\u7684\u8f49\u63db\u51fd\u6578\u90fd\u662f\u57fa\u65bc\u6536\u96c6\u5927\u91cf\u7684\u4e7e\u6de8\u53ca\u96dc\u8a0a\u8a9e\u97f3\u8cc7\u6599\u8a13\u7df4\u800c\u5f97\u3002 \u64c7\u70ba\u8eca\u566a\u97f3\u548c\u5b30\u5152\u54ed\u8072\u3002\u672c\u7814\u7a76\u7684\u76ee\u7684\u5728\u65bc\u958b\u767c\u4f7f\u7528\u8005\u4ecb\u9762\uff0c\u56e0\u6b64\u5728\u6a5f\u5668\u5b78\u7fd2\u65b9\u6cd5\u7684\u6e2c DDAE</td></tr><tr><td>\u8a9e\u97f3\u589e\u5f37\u6280\u8853\u70ba\u5404\u9805\u8a9e\u97f3\u8a0a\u865f\u6280\u8853\u4e4b\u91cd\u8981\u524d\u8655\u7406\u55ae\u5143\uff0c\u91dd\u5c0d\u6536\u96c6\u5230\u7684\u8072\u97f3\u8a0a\u865f\u6291\u5236 \u74b0\u5883\u566a\u97f3\u4f86\u589e\u5f37\u8a0a\u865f\u7684\u54c1\u8cea\uff0c\u9032\u800c\u63d0\u5347\u5404\u9805\u61c9\u7528\u7684\u6548\u80fd\u3002\u7136\u800c\u4e0d\u540c\u7684\u8a9e\u97f3\u589e\u5f37\u6280\u8853\u6709\u4e0d \u96dc\u6bd4 (SNR)\uff0c\u6211\u5011\u8a2d\u8a08\u4e00\u500b\u6620\u5c04\u51fd\u6578(\u5982\u5716\u4e00)\u4f86\u6c7a\u5b9a\u6700\u4f73\u7684\u4e8b\u524d\u6a5f\u7387\u6bd4\u4f8b\u3002 \u591a\u6a5f\u5668\u5b78\u7fd2\u7406\u8ad6\u4e2d\uff0c\u53c8\u4ee5\u6df1\u5c64\u5b78\u7fd2\u7406\u8ad6(deep learning)\u6700\u70ba\u53d7\u5230\u77da\u76ee\u3002\u76f8\u8f03\u65bc\u50b3\u7d71\u7684\u6a5f\u5668 \u5b78\u7fd2\u7406\u8ad6\uff0c\u6df1\u5c64\u5b78\u7fd2\u7406\u8ad6\u5229\u7528\u591a\u5c64\u5f0f\u7d50\u69cb\u67b6\u69cb\u51fa\u4e00\u500b\u975e\u7dda\u6027\u4e14\u8907\u96dc\u7684\u6a21\u578b\uff0c\u5c0d\u65bc\u591a\u9805\u6a19 \u6b64\u7814\u7a76\u6240\u958b\u767c\u7684\u8a9e\u97f3\u589e\u5f37\u4ecb\u9762\u5305\u542b\u4e0a\u8ff0\u7684\u50b3\u7d71\u65b9\u6cd5\u8207\u6a5f\u5668\u5b78\u7fd2\u964d\u566a\u65b9\u6cd5\uff0c\u4ecb\u9762\u5982\u5716 \u505a\u70ba\u5ba2\u89c0\u8a55\u91cf(\u5982\u8868\u4e8c)\uff0c\u5f9e\u5ba2\u89c0\u8a55\u91cf\u7684\u7d50\u679c\u53ef\u4ee5\u89c0\u5bdf\u51fa\u50b3\u7d71\u65b9\u6cd5\u8207\u6a5f\u5668\u5b78\u7fd2\u65b9\u6cd5\u5728 \u4e09\uff0c\u4f7f\u7528\u8005\u53ef\u5728\u5716\u4e09\u2460\u2461\u8f38\u5165\u6b32\u8655\u7406\u7684\u8a9e\u97f3\u8207\u52a0\u6210\u7684\u566a\u97f3\uff0c\u5728\u2462\u8f38\u5165\u5408\u6210\u7684 SNR \u9032 \u5152\u54ed\u8072\u3002\u5f9e\u8868\u4e00\u7684\u7d50\u679c\u53ef\u767c\u73fe\u50b3\u7d71\u8207\u6a5f\u5668\u5b78\u7fd2\u65b9\u6cd5\u90fd\u80fd\u6709\u6548\u6291\u5236\u8eca\u566a\u97f3\u3002\u6211\u5011\u9078\u7528 PESQ \u8868\u56db \u5404\u8a9e\u97f3\u589e\u5f37\u65b9\u6cd5\u5c0d\u5b30\u5152\u54ed\u8072\u8655\u7406\u5f8c\u7684 PESQ</td></tr><tr><td>\u540c\u7684\u61c9\u7528\uff0c\u50b3\u7d71\u7684\u8a9e\u97f3\u589e\u5f37\u6f14\u7b97\u6cd5\u9069\u5408\u8655\u7406\u7a69\u614b\u7684\u74b0\u5883\u566a\u97f3\uff0c\u8b6c\u5982\u8eca\u8072\u3001\u5de5\u5ee0\u8072\u7b49\u80fd\u91cf \u6e96\u7684\u8a0a\u865f\u8655\u7406\u3001\u6a21\u5f0f\u8b58\u5225\u7b49\u6e2c\u8a66\u9805\u76ee\uff0c\u5df2\u6709\u8af8\u591a\u512a\u7570\u7684\u7814\u7a76\u6210\u679c\u8207\u8868\u73fe\uff0c\u751a\u81f3\u662f\u9019\u4e9b\u9818 \u884c\u6df7\u566a\uff0c\u82e5\u4e0d\u6df7\u566a\u53ef\u8f38\u5165'N'\uff0c\u6df7\u566a\u5f8c\u7684\u8a9e\u97f3\u6ce2\u5f62\u5716\u8207\u8072\u8b5c\u5716\u5c07\u986f\u793a\u5728\u2465\u3002\u5728\u8f38\u5165\u8a9e\u97f3 PESQ \u4e0a\u7684\u8868\u73fe\u7121\u986f\u8457\u7684\u5dee\u7570\u3002</td></tr><tr><td>\u8207\u566a\u97f3\u6642\uff0c\u4ecb\u9762\u6703\u986f\u793a\u8f38\u5165\u8a0a\u865f\u7684\u983b\u7387\u8cc7\u8a0a\uff0c\u82e5\u5169\u8005\u7684\u983b\u7387\u4e0d\u540c\u5247\u7121\u6cd5\u9032\u884c\u6df7\u566a\u8655\u7406\u3002</td></tr><tr><td>\u96c6\u4e2d\u5728\u67d0\u4e9b\u983b\u7387\u7684\u566a\u97f3\uff0c\u4f46\u975e\u7a69\u614b\u7684\u566a\u97f3\u5982\u9cf4\u7b1b\u8072\u3001\u4eba\u8072\u3001\u98a8\u5207\u8072\u7b49\u5247\u662f\u4ee5\u6a5f\u5668\u5b78\u7fd2\u6240 \u57df\u6700\u5148\u9032\u7684\u6280\u8853\u3002\u5716\u4e8c\u70ba\u6d41\u7a0b\u5716\uff0c\u57fa\u65bc\u6df1\u5c64\u5b78\u7fd2\u7406\u8ad6\u7684\u6df1\u5c64\u53bb\u566a\u81ea\u7de8\u78bc\u6a21\u578b\u61c9\u7528\u65bc\u8a9e\u97f3 \u8868\u4e00 \u5404\u8a9e\u97f3\u589e\u5f37\u65b9\u6cd5\u5c0d\u8eca\u566a\u97f3\u7684\u8655\u7406\u6210\u6548\u3002</td></tr><tr><td>\u767c\u5c55\u7684\u6f14\u7b97\u6cd5\u8f03\u70ba\u6709\u6548\uff0c\u56e0\u6b64\u91dd\u5c0d\u4e0d\u540c\u74b0\u5883\u6216\u4f7f\u7528\u9700\u6c42\u61c9\u4f7f\u7528\u4e0d\u540c\u7684\u8655\u7406\u65b9\u5f0f\u3002\u672c\u7814\u7a76 \u589e\u5f37\u6280\u8853\u3002 \u964d\u566a\u7684\u65b9\u6cd5\u5206\u70ba\u5169\u90e8\u5206\uff0c\u5728\u2463\u9078\u64c7\u566a\u97f3\u4f30\u6e2c\u6a21\u578b\u4e26\u65bc\u2464\u9078\u64c7\u8a9e\u97f3\u589e\u5f37\u65b9\u6cd5\u3002\u8655\u7406\u5f8c\u7684 \u8868\u4e09\u7684\u566a\u97f3\u70ba\u5b30\u5152\u54ed\u8072\uff0c\u5f9e\u8868\u4e09\u53ef\u770b\u51fa\u50b3\u7d71\u7684\u65b9\u5f0f\u4e26\u7121\u6cd5\u6709\u6548\u7684\u964d\u4f4e\u5b30\u5152\u54ed\u8072\uff0c\u800c</td></tr><tr><td>\u9810\u8a08\u5efa\u7acb\u4f7f\u7528\u8005\u4ecb\u9762\uff0c\u5176\u4f7f\u7528\u8005\u4ecb\u9762\u5305\u542b\u50b3\u7d71\u8207\u6a5f\u5668\u5b78\u7fd2\u7684\u8a9e\u97f3\u589e\u5f37\u65b9\u6cd5\uff1b\u671f\u671b\u4e00\u5957\u7c21 \u7d50\u679c\u5982\u5716\u4e09\uff0c\u5728\u2466\u8207\u2467\u5206\u5225\u986f\u793a\u6642\u57df\u8a9e\u97f3\u6ce2\u5f62\u5716\u8207\u8072\u8b5c\u8cc7\u8a0a\u3002\u6b64\u4ecb\u9762\u4e5f\u63d0\u4f9b\u4e86 NMF \u8655\u7406\u524d\u8a9e\u97f3 \u6a5f\u5668\u5b78\u7fd2\u7684\u8a9e\u97f3\u589e\u5f37\u65b9\u6cd5\u53ef\u4ee5\u6709\u6548\u7684\u964d\u4f4e\u5b30\u5152\u54ed\u8072\u3002\u5b30\u5152\u54ed\u8072\u70ba\u975e\u7a69\u614b\u7684\u566a\u97f3\uff0c\u4e14\u5b30\u5152</td></tr><tr><td>\u55ae\u7684\u4f7f\u7528\u8005\u4ecb\u9762\u53ef\u4ee5\u63d0\u4f9b\u4f7f\u7528\u8005\u4ee5\u8f03\u4fbf\u5229\u7684\u65b9\u5f0f\u9032\u884c\u591a\u7a2e\u8a9e\u97f3\u589e\u5f37\u65b9\u6cd5\u6a21\u64ec\u8207\u6bd4\u8f03\uff0c\u8f14 \u8207 DDAE \u6a21\u578b\u8a13\u7df4\u529f\u80fd\uff0c\u4f7f\u7528\u8005\u53ef\u4ee5\u81ea\u7531\u8a2d\u5b9a\u53c3\u6578\u548c\u9078\u64c7\u8a13\u7df4\u8cc7\u6599\u5373\u53ef\u751f\u6210\u6a21\u578b\u3002 \u54ed\u8072\u7684\u8072\u5b78\u7279\u5fb5\u985e\u4f3c\u4eba\u8072\uff0c\u4f7f\u7528\u50b3\u7d71\u65b9\u6cd5\u8f03\u96e3\u6e96\u78ba\u7684\u9810\u4f30\u4e26\u6d88\u9664\u3002\u5f9e\u5ba2\u89c0\u8a55\u91cf\u4e2d(\u5982\u8868</td></tr><tr><td>\u97cb\u7d0d\u6ffe\u6ce2\u5668 \u52a9\u4f7f\u7528\u8005\u5feb\u901f\u7684\u9078\u64c7\u9069\u5408\u7684\u8a9e\u97f3\u589e\u5f37\u65b9\u6cd5\u3002 \u56db)\u4e5f\u53ef\u5f97\u5230\u76f8\u540c\u7684\u7d50\u8ad6\u3002</td></tr><tr><td>\u8868\u4e09 \u5404\u8a9e\u97f3\u589e\u5f37\u65b9\u5f0f\u5c0d\u5b30\u5152\u54ed\u8072\u7684\u8655\u7406\u6210\u6548\u3002</td></tr><tr><td>\u4e8c\u3001\u7406\u8ad6</td></tr><tr><td>\u50b3\u7d71\u8a9e\u97f3\u589e\u5f37\u65b9\u6cd5 KLT \u8655\u7406\u524d\u8a9e\u97f3</td></tr><tr><td>\u5716\u4e00 GMAPA \u6620\u5c04\u51fd\u6578\u793a\u610f\u5716\u3002</td></tr><tr><td>\u50b3\u7d71\u7684\u8a9e\u97f3\u589e\u5f37\u7cfb\u7d71\u4e3b\u8981\u76ee\u7684\u70ba\u6d88\u9664\u80cc\u666f\u566a\u97f3\u53ca\u964d\u4f4e\u589e\u5f37\u5f8c\u8a9e\u97f3\u8a0a\u865f\u7684\u5931\u771f\u3002\u5927\u591a</td></tr><tr><td>\u6a5f\u5668\u5b78\u7fd2\u8a9e\u97f3\u589e\u5f37\u65b9\u6cd5</td></tr><tr><td>\u6578\u7684\u8a9e\u97f3\u589e\u5f37\u6280\u8853\u5728\u983b\u57df\u4e0a\u5f37\u5316\u8072\u97f3\u8a0a\u865f\uff0c\u901a\u5e38\u7531\u5169\u500b\u5b50\u7cfb\u7d71\u7d50\u5408\u800c\u6210\uff0c\u5206\u5225\u662f\u96dc\u8a0a\u8207 MMSE \u97cb\u7d0d\u6ffe\u6ce2\u5668</td></tr><tr><td>\u589e\u76ca\u503c\u4f30\u6e2c\u7cfb\u7d71[1]\u3002\u9996\u5148\u85c9\u7531\u77ed\u6642\u5085\u5229\u8449\u8f49\u63db\u5c07\u8a0a\u865f\u9032\u884c\u5206\u983b\u7684\u8655\u7406\uff0c\u4e26\u53d6\u5f97\u5e36\u566a\u8a9e \u50b3\u7d71\u7684\u8a9e\u97f3\u589e\u5f37\u662f\u91dd\u5c0d\u8a0a\u865f\u9032\u884c\u566a\u97f3\u8207\u8a0a\u865f\u7684\u9810\u4f30\uff0c\u4f46\u5c0d\u65bc\u975e\u7a69\u614b\u7684\u566a\u97f3(\u4f8b\u5982:\u8b66</td></tr><tr><td>\u97f3\u983b\u8b5c\u4e0a\u7684\u632f\u5e45\u8207\u76f8\u4f4d\u3002\u4fdd\u7559\u76f8\u4f4d\u6210\u4efd\uff0c\u96dc\u8a0a\u8207\u589e\u76ca\u503c\u4f30\u6e2c\u7cfb\u7d71\u5f37\u5316\u632f\u5e45\u6210\u4efd\uff0c\u6700\u5f8c\u7d93 \u7b1b\u8072\u3001\u4eba\u8072\u7b49)\u964d\u566a\u6548\u679c\u8f03\u5dee\u3002\u8fd1\u5e7e\u5e74\u5feb\u901f\u767c\u5c55\u7684\u6a5f\u5668\u5b78\u7fd2\u4e2d\uff0cDDAE (deep denoising</td></tr><tr><td>\u53cd\u77ed\u6642\u5085\u5229\u8449\u8f49\u63db\u5c07\u5176\u91cd\u7d44\u70ba\u6642\u57df\u4e0a\u8f03\u70ba\u4e7e\u6de8\u7684\u8072\u97f3\u8a0a\u865f\u3002\u50b3\u7d71\u7684\u65b9\u6cd5\u4e2d\uff0c\u8f03\u5e38\u898b\u7684\u8a9e auto-encoder) [9][10] \u548c NMF (non-negative matrix factorization) [11-13]\u5169\u7a2e\u6280\u8853\u4e5f\u5e38\u88ab GMAPA KLT</td></tr><tr><td>\u97f3\u589e\u5f37\u65b9\u5f0f\u5305\u4e86\u97cb\u7d0d\u6ffe\u6ce2\u5668[2][3]\u3001\u983b\u8b5c\u522a\u6e1b\u6cd5[4]\u3001\u6700\u5c0f\u5316\u5747\u65b9\u8aa4\u5dee\u4f30\u6e2c [5]\u548c\u6700\u5927\u4e8b \u61c9\u7528\u65bc\u8a0a\u865f\u589e\u5f37\u7684\u9818\u57df\u4e2d\u3002\u975e\u8ca0\u77e9\u9663\u5206\u89e3\u6280\u8853\u85c9\u7531\u57fa\u5e95\u77e9\u9663 W \u8207\u7de8\u78bc\u77e9\u9663 H \u76f8\u4e58\u4ee5\u8fd1 \u5716\u4e8c \u61c9\u7528\u6df1\u5c64\u53bb\u566a\u81ea\u7de8\u78bc\u6a21\u578b\u65bc\u8a9e\u97f3\u589e\u5f37\u3002</td></tr><tr><td>\u5f8c\u983b\u8b5c\u632f\u5e45\u9810\u4f30\u5668\u7684\u8a9e\u97f3\u589e\u5f37\u6f14\u7b97\u6cd5(Generalized Maximum A Posteriori spectral Amplitude, GMAPA)[6]\u7b49\u3002 \u4f3c\u8f38\u5165\u983b\u8b5c V\uff0c\u5982\u5f0f(1)\uff1a V\u2248WH, \u89e3\u6642\u57df\u8a0a\u865f\u81f3\u5176\u983b\u8b5c\u6210\u4efd\u4e26\u4fdd\u7559\u76f8\u4f4d\u8cc7\u8a0a\u3002\u632f\u5e45\u8cc7\u8a0a\u8f38\u5165\u6df1\u5c64\u53bb\u566a\u81ea\u7de8\u78bc\u6a21\u578b\u964d\u566a\uff0c\u7372 (1) \u5982\u5716\u4e8c\u6240\u793a\uff0c\u985e\u4f3c\u65bc\u50b3\u7d71\u8a9e\u97f3\u589e\u5f37\u6280\u8853\uff0c\u5e36\u566a\u8a9e\u97f3\u8a0a\u865f\u9996\u5148\u7d93\u7531\u77ed\u6642\u5085\u5229\u8449\u8f49\u63db\u62c6 \u5716\u4e09 \u4f7f\u7528\u8005\u4ecb\u9762\u793a\u610f\u5716\u3002 NMF MMSE</td></tr><tr><td>\u97cb\u7d0d\u6ffe\u6ce2\u5668\u3001\u983b\u8b5c\u522a\u6e1b\u6cd5\u7b49\u5178\u578b\u7684\u8a9e\u97f3\u589e\u5f37\u65b9\u6cd5\u5df2\u5ee3\u88ab\u61c9\u7528\uff0c\u904e\u53bb\u6709\u8a31\u591a\u7814\u7a76\u6539 \u5f97\u8f03\u70ba\u4e7e\u6de8\u7684\u632f\u5e45\u983b\u8b5c\uff0c\u6700\u5f8c\u7531\u53cd\u77ed\u6642\u5085\u5229\u8449\u8f49\u63db\u5c07\u8f03\u70ba\u4e7e\u6de8\u7684\u632f\u5e45\u983b\u8b5c\u8207\u76f8\u4f4d\u91cd\u5efa\u70ba</td></tr></table>", |
| "type_str": "table" |
| } |
| } |
| } |
| } |
