| { | |
| "paper_id": "O17-1031", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T07:59:35.238979Z" | |
| }, | |
| "title": "", | |
| "authors": [], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "", | |
| "pdf_parse": { | |
| "paper_id": "O17-1031", | |
| "_pdf_hash": "", | |
| "abstract": [], | |
| "body_text": [ | |
| { | |
| "text": ") ; ( ) ( ) , ( x t h t s x t y t coch \uf0c4 \uf03d (2.1) ) ( )) , ( ( ) , ( t w x t y g x t y t coch t AN \uf0c4 \uf0b6 \uf03d (2.2) ) 0 ), , ( max( ) , ( x t y x t y AN x LIN \uf0b6 \uf03d (2.3) ) ; ( ) , ( ) , ( \uf074 t u x t y x t y t LIN final \uf0c4 \uf03d (2.4) \u5f0f(2,1)\u70ba\u5206\u6790\u90e8\u5206\uff0c\u7528\uf92d\u6a21\u64ec\u8072\u97f3 ) (t s \u50b3\u81f3\u8033\u8778\u5f8c\uff0c\u5728\u57fa\u5e95\u819c\u4e0a\u4f9d\u7167\u5176\u672c\u8eab\uf967\u540c\u7684\u5171\u632f\u983b \uf961\uff0c\u88ab\uf967\u540c\u7684\u4f4d\u7f6e\u4e0a\u88ab\u89e3\u6790\u51fa\uf92d\u3002 ) , ( x t h \u4ee3\u8868\u57fa\u5e95\u819c\u4e0a\u4f4d\u7f6e\u70ba x \u7684\u5171\u632f\u97ff\u61c9\uff0c", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| } | |
| ], | |
| "back_matter": [], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "A unique approach in text independent speaker recognition using MFCC feature sets and probabilistic neural network", | |
| "authors": [ | |
| { | |
| "first": "Anil", | |
| "middle": [ | |
| "S" | |
| ], | |
| "last": "Khan Suhail Ahmad", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jagannath", | |
| "middle": [ | |
| "H" | |
| ], | |
| "last": "Thosar", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Vinay", | |
| "middle": [ | |
| "S" | |
| ], | |
| "last": "Nirmal", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Pande", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2015, | |
| "venue": "Proc. of Advances in Pattern Recognition (ICAPR)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "1--6", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Khan Suhail Ahmad, Anil S. Thosar, Jagannath H. Nirmal, and Vinay S. Pande, \"A unique approach in text independent speaker recognition using MFCC feature sets and probabilistic neural network,\" in Proc. of Advances in Pattern Recognition (ICAPR), pp. 1-6, 2015.", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "Speaker recognition based on MFCC and BP neural networks", | |
| "authors": [ | |
| { | |
| "first": "Yi", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Bob", | |
| "middle": [], | |
| "last": "Lawlor", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2017, | |
| "venue": "Proc. of Signals and Systems Conference (ISSC)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "1--4", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Yi Wang, and Bob Lawlor, \"Speaker recognition based on MFCC and BP neural networks,\" in Proc. of Signals and Systems Conference (ISSC), pp. 1-4, 2017.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Deep neural networks for cochannel speaker identification", | |
| "authors": [ | |
| { | |
| "first": "Xiaojia", | |
| "middle": [], | |
| "last": "Zhao", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Yuxuan", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Deliang", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2015, | |
| "venue": "Proc. of ICASSP", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "4824--4828", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Xiaojia Zhao, Yuxuan Wang, and DeLiang Wang, \"Deep neural networks for cochannel speaker identification,\" in Proc. of ICASSP, pp. 4824-4828, 2015.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "Speech acoustic modeling from raw multichannel waveforms", | |
| "authors": [ | |
| { | |
| "first": "Yedid", | |
| "middle": [], | |
| "last": "Hoshen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ron", | |
| "middle": [ | |
| "J" | |
| ], | |
| "last": "Weiss", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Kevin", | |
| "middle": [ | |
| "W" | |
| ], | |
| "last": "Wilson", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2015, | |
| "venue": "Proc. of ICASSP", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "4624--4628", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Yedid Hoshen, Ron J. Weiss, and Kevin W. Wilson, \"Speech acoustic modeling from raw multichannel waveforms,\" in Proc. of ICASSP, pp. 4624-4628, 2015.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "Very deep convolutional neural networks for raw waveforms", | |
| "authors": [ | |
| { | |
| "first": "Wei", | |
| "middle": [], | |
| "last": "Dai", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Chia", | |
| "middle": [], | |
| "last": "Dai", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Shuhui", | |
| "middle": [], | |
| "last": "Qu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Juncheng", | |
| "middle": [], | |
| "last": "Li", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Samarjit", | |
| "middle": [], | |
| "last": "Das", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2017, | |
| "venue": "Proc. of ICASSP", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "421--425", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Wei Dai, Chia Dai, Shuhui Qu, Juncheng Li, and Samarjit Das, \"Very deep convolutional neural networks for raw waveforms,\" in Proc. of ICASSP, pp. 421-425, 2017.", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "Learning the speech front-end with raw waveform CLDNNs", | |
| "authors": [ | |
| { | |
| "first": "Tara", | |
| "middle": [ | |
| "N" | |
| ], | |
| "last": "Sainath", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ron", | |
| "middle": [ | |
| "J" | |
| ], | |
| "last": "Weiss", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Andrew", | |
| "middle": [], | |
| "last": "Senior", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Kevin", | |
| "middle": [ | |
| "W" | |
| ], | |
| "last": "Wilson", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Oriol", | |
| "middle": [], | |
| "last": "Vinyals", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2015, | |
| "venue": "Proc. of INTERSPEECH", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "1--5", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Tara N. Sainath, Ron J. Weiss, Andrew Senior, Kevin W. Wilson, and Oriol Vinyals, \"Learning the speech front-end with raw waveform CLDNNs,\" in Proc. of INTERSPEECH, pp. 1-5, 2015.", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "An information theoretical investigation into the distribution of phonetic information across the auditory spectrogram", | |
| "authors": [ | |
| { | |
| "first": "Andrew", | |
| "middle": [], | |
| "last": "Morris", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jean-Luc", | |
| "middle": [], | |
| "last": "Schwartz", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Pierre", | |
| "middle": [], | |
| "last": "Escudier", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1993, | |
| "venue": "Computer Speech & Language", | |
| "volume": "7", | |
| "issue": "", | |
| "pages": "121--136", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Andrew Morris, Jean-Luc Schwartz, and Pierre Escudier, \"An information theoretical investigation into the distribution of phonetic information across the auditory spectrogram,\" Computer Speech & Language 7.2: 121-136, 1993", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "Speech-recognition difficulties of the hearing-impaired elderly: The contributions of audibility", | |
| "authors": [ | |
| { | |
| "first": "Larry", | |
| "middle": [ | |
| "E" | |
| ], | |
| "last": "Humes", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Lisa", | |
| "middle": [], | |
| "last": "Roberts", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1990, | |
| "venue": "Journal of Speech, Language, and Hearing Research", | |
| "volume": "33", | |
| "issue": "", | |
| "pages": "726--735", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Larry E. Humes, and Lisa Roberts, \"Speech-recognition difficulties of the hearing-impaired elderly: The contributions of audibility,\" Journal of Speech, Language, and Hearing Research, 33.4: 726-735, 1990.", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "Perceptual consequences of cochlear hearing loss and their implications for the design of hearing aids", | |
| "authors": [ | |
| { | |
| "first": "C", | |
| "middle": [ | |
| "J" | |
| ], | |
| "last": "Brian", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Moore", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1996, | |
| "venue": "Ear and hearing", | |
| "volume": "17", | |
| "issue": "", | |
| "pages": "133--161", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Brian C. J. Moore, \"Perceptual consequences of cochlear hearing loss and their implications for the design of hearing aids,\" Ear and hearing, 17.2: 133-161, 1996.", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "Multi-resolution spectro-temporal analysis of complex sounds", | |
| "authors": [ | |
| { | |
| "first": "T", | |
| "middle": [], | |
| "last": "Chi", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "P", | |
| "middle": [], | |
| "last": "Ru", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "S", | |
| "middle": [ | |
| "A" | |
| ], | |
| "last": "Shamma", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": null, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "T. Chi, P. Ru, and S. A. Shamma, \"Multi-resolution spectro-temporal analysis of complex sounds,\" J.", | |
| "links": null | |
| }, | |
| "BIBREF12": { | |
| "ref_id": "b12", | |
| "title": "A spectro-temporal modulation index (stmi) for assessment of speech intelligibility", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Elhilali", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "T", | |
| "middle": [], | |
| "last": "Chi", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Shamma", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2003, | |
| "venue": "Speech Communication", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "331--348", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "M. Elhilali, T. Chi, and S. Shamma, \"A spectro-temporal modulation index (stmi) for assessment of speech intelligibility,\" Speech Communication, pp. 331-348, 2003.", | |
| "links": null | |
| }, | |
| "BIBREF13": { | |
| "ref_id": "b13", | |
| "title": "Spectro-temporal modulation energy based mask for robust speaker identification", | |
| "authors": [ | |
| { | |
| "first": "T.-S", | |
| "middle": [], | |
| "last": "Chi", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "T.-H", | |
| "middle": [], | |
| "last": "Lin", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "C.-C", | |
| "middle": [], | |
| "last": "Hsu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2012, | |
| "venue": "J. Acoust. Soc. Am", | |
| "volume": "131", | |
| "issue": "5", | |
| "pages": "368--374", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "T.-S. Chi, T.-H. Lin, and C.-C. Hsu, \"Spectro-temporal modulation energy based mask for robust speaker identification,\" J. Acoust. Soc. Am., vol. 131, no. 5, pp. EL368-EL374, 2012.", | |
| "links": null | |
| }, | |
| "BIBREF14": { | |
| "ref_id": "b14", | |
| "title": "Spectro-temporal modulation based singing detection combined with pitch-based grouping for singing voice separation", | |
| "authors": [ | |
| { | |
| "first": "T", | |
| "middle": [ | |
| "E" | |
| ], | |
| "last": "Lin", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "C", | |
| "middle": [ | |
| "C" | |
| ], | |
| "last": "Hsu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y", | |
| "middle": [ | |
| "C" | |
| ], | |
| "last": "Chen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [ | |
| "H" | |
| ], | |
| "last": "Chen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "T", | |
| "middle": [ | |
| "S" | |
| ], | |
| "last": "Chi", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2013, | |
| "venue": "Proc. of INTERSPEECH", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "2920--2923", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "T. E. Lin, C. C. Hsu, Y. C. Chen, J. H. Chen, and T. S. Chi, \"Spectro-temporal modulation based singing detection combined with pitch-based grouping for singing voice separation,\" in Proc. of INTERSPEECH., pp. 2920-2923, 2013.", | |
| "links": null | |
| }, | |
| "BIBREF15": { | |
| "ref_id": "b15", | |
| "title": "Singing voice separation using spectro-temporal modulation features", | |
| "authors": [ | |
| { | |
| "first": "F", | |
| "middle": [], | |
| "last": "Yen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y.-J", | |
| "middle": [], | |
| "last": "Luo", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "T.-S", | |
| "middle": [], | |
| "last": "Chi", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "Proc. of Annual Conference of International Society for Music Information Retrieval (ISMIR)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "617--622", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "F. Yen, Y.-J. Luo, and T.-S. Chi, \"Singing voice separation using spectro-temporal modulation features,\" in Proc. of Annual Conference of International Society for Music Information Retrieval (ISMIR), pp. 617-622, 2014.", | |
| "links": null | |
| }, | |
| "BIBREF16": { | |
| "ref_id": "b16", | |
| "title": "Auditory attention-focusing the searchlight on sound", | |
| "authors": [ | |
| { | |
| "first": "J", | |
| "middle": [ | |
| "B" | |
| ], | |
| "last": "Fritz", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Elhilali", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "S", | |
| "middle": [ | |
| "V" | |
| ], | |
| "last": "David", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "S", | |
| "middle": [ | |
| "A" | |
| ], | |
| "last": "Shamma", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2007, | |
| "venue": "Current opinion in neurobiology", | |
| "volume": "17", | |
| "issue": "4", | |
| "pages": "437--455", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "J. B. Fritz, M. Elhilali, S. V. David, and S. A. Shamma, \"Auditory attention-focusing the searchlight on sound,\" Current opinion in neurobiology, vol. 17, no. 4, pp. 437-455, 2007.", | |
| "links": null | |
| }, | |
| "BIBREF17": { | |
| "ref_id": "b17", | |
| "title": "Auditory perception of sound sources", | |
| "authors": [ | |
| { | |
| "first": "E", | |
| "middle": [ | |
| "R" | |
| ], | |
| "last": "Hafter", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Sarampalis", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "P", | |
| "middle": [], | |
| "last": "Loui", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2008, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "115--142", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "E. R. Hafter, A. Sarampalis, and P. Loui, \"Auditory attention and filters,\" Auditory perception of sound sources, Springer US, pp. 115-142, 2008.", | |
| "links": null | |
| }, | |
| "BIBREF18": { | |
| "ref_id": "b18", | |
| "title": "Auditory cortical receptive fields: Stable entities with plastic abilities", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Elhilali", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Fritz", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "T", | |
| "middle": [], | |
| "last": "Chi", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Shamma", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2007, | |
| "venue": "J. Neuroscience", | |
| "volume": "27", | |
| "issue": "39", | |
| "pages": "10--372", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "M. Elhilali, J. Fritz, T. Chi, and S. Shamma, \"Auditory cortical receptive fields: Stable entities with plastic abilities,\" J. Neuroscience, vol. 27, no. 39, pp. 10 372-10 382, 2007.", | |
| "links": null | |
| }, | |
| "BIBREF19": { | |
| "ref_id": "b19", | |
| "title": "A joint training framework for robust automatic speech recognition", | |
| "authors": [ | |
| { | |
| "first": "Z", | |
| "middle": [ | |
| "Q" | |
| ], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": null, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Z. Q. Wang and D. Wang, \"A joint training framework for robust automatic speech recognition,\"", | |
| "links": null | |
| }, | |
| "BIBREF21": { | |
| "ref_id": "b21", | |
| "title": "Learning acoustic frame labeling for speech recognition with recurrent neural networks", | |
| "authors": [ | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Sak", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Senior", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "K", | |
| "middle": [], | |
| "last": "Rao", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "O", | |
| "middle": [], | |
| "last": "Irsoy", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Graves", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "F", | |
| "middle": [], | |
| "last": "Beaufays", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Schalkwyk", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2015, | |
| "venue": "Proc. of ICASSP", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "4280--4284", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "H. Sak, A. Senior, K. Rao, O. Irsoy, A. Graves, F. Beaufays, and J. Schalkwyk, \"Learning acoustic frame labeling for speech recognition with recurrent neural networks,\" in Proc. of ICASSP, pp. 4280-4284, 2015.", | |
| "links": null | |
| }, | |
| "BIBREF22": { | |
| "ref_id": "b22", | |
| "title": "Complex ratio masking for monaural speech separation", | |
| "authors": [ | |
| { | |
| "first": "D", | |
| "middle": [ | |
| "S" | |
| ], | |
| "last": "Williamson", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": null, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "D. S. Williamson, Y. Wang, and D. Wang, \"Complex ratio masking for monaural speech separation,\"", | |
| "links": null | |
| }, | |
| "BIBREF24": { | |
| "ref_id": "b24", | |
| "title": "Spectro-temporal modulations for robust speech emotion recognition", | |
| "authors": [ | |
| { | |
| "first": "L.-Y", | |
| "middle": [], | |
| "last": "Yeh", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "T.-S", | |
| "middle": [], | |
| "last": "Chi", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2010, | |
| "venue": "Proc. of INTERSPEECH", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "789--792", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "L.-Y. Yeh and T.-S. Chi, \"Spectro-temporal modulations for robust speech emotion recognition,\" in Proc. of INTERSPEECH, pp. 789-792, 2010.", | |
| "links": null | |
| }, | |
| "BIBREF25": { | |
| "ref_id": "b25", | |
| "title": "Imagenet classification with deep convolutional neural networks", | |
| "authors": [ | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Krizhevsky", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "I", | |
| "middle": [], | |
| "last": "Sutskever", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "G", | |
| "middle": [ | |
| "E" | |
| ], | |
| "last": "Hinton", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2012, | |
| "venue": "Advances in neural information processing systems", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "1097--1105", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "A. Krizhevsky, I. Sutskever, and G. E. Hinton, \"Imagenet classification with deep convolutional neural networks,\" Advances in neural information processing systems, pp. 1097-1105, 2012.", | |
| "links": null | |
| }, | |
| "BIBREF26": { | |
| "ref_id": "b26", | |
| "title": "Stacked convolutional auto-encoders for hierarchical feature extraction", | |
| "authors": [ | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Masci", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "U", | |
| "middle": [], | |
| "last": "Meier", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Cirean", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Schmidhuber", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2011, | |
| "venue": "Proc. of International Conference on Artificial Neural Networks", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "52--59", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "J. Masci, U. Meier, D. Cirean, and J. Schmidhuber, \"Stacked convolutional auto-encoders for hierarchical feature extraction,\" In Proc. of International Conference on Artificial Neural Networks, pp. 52-59, 2011.", | |
| "links": null | |
| }, | |
| "BIBREF27": { | |
| "ref_id": "b27", | |
| "title": "Convolutional, long short-term memory, fully connected deep neural networks", | |
| "authors": [ | |
| { | |
| "first": "T", | |
| "middle": [ | |
| "N" | |
| ], | |
| "last": "Sainath", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "O", | |
| "middle": [], | |
| "last": "Vinyals", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Senior", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Sak", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2015, | |
| "venue": "Proc. of ICASSP", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "4580--4584", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "T. N. Sainath, O. Vinyals, A. Senior, and H. Sak, \"Convolutional, long short-term memory, fully connected deep neural networks.\" in Proc. of ICASSP, pp. 4580-4584, 2015.", | |
| "links": null | |
| }, | |
| "BIBREF28": { | |
| "ref_id": "b28", | |
| "title": "Applying convolutional neural networks concepts to hybrid nn-hmm model for speech recognition", | |
| "authors": [ | |
| { | |
| "first": "O", | |
| "middle": [], | |
| "last": "Hamid", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [ | |
| "R" | |
| ], | |
| "last": "Mohamed", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Jiang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "G", | |
| "middle": [], | |
| "last": "Penn", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2012, | |
| "venue": "Proc. of ICASSP", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "4277--4280", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "O. Abdel-Hamid, A. R. Mohamed, H. Jiang, and G. Penn, \"Applying convolutional neural networks concepts to hybrid nn-hmm model for speech recognition.\" in Proc. of ICASSP, pp. 4277-4280, 2012.", | |
| "links": null | |
| }, | |
| "BIBREF29": { | |
| "ref_id": "b29", | |
| "title": "Effect of enhancement of spectral changes on speech intelligibility and clarity preferences for the hearing impaired", | |
| "authors": [ | |
| { | |
| "first": "Jing", | |
| "middle": [], | |
| "last": "Chen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Thomas", | |
| "middle": [], | |
| "last": "Baer", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Brian Cj", | |
| "middle": [], | |
| "last": "Moore", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2012, | |
| "venue": "J. Acoust. Soc. Am", | |
| "volume": "131", | |
| "issue": "", | |
| "pages": "2987--2998", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Jing Chen, Thomas Baer, and Brian CJ Moore. \"Effect of enhancement of spectral changes on speech intelligibility and clarity preferences for the hearing impaired.\" J. Acoust. Soc. Am., 131.4: 2987-2998, 2012", | |
| "links": null | |
| }, | |
| "BIBREF30": { | |
| "ref_id": "b30", | |
| "title": "Multiband analysis and synthesis of spectro-temporal modulations of Fourier spectrogram", | |
| "authors": [ | |
| { | |
| "first": "Tai-Shih", | |
| "middle": [], | |
| "last": "Chi", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Chung-Chien", | |
| "middle": [], | |
| "last": "Hsu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2011, | |
| "venue": "J. Acoust. Soc. Am., 129", | |
| "volume": "5", | |
| "issue": "", | |
| "pages": "190--196", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Tai-Shih Chi, and Chung-Chien Hsu. \"Multiband analysis and synthesis of spectro-temporal modulations of Fourier spectrogram.\" J. Acoust. Soc. Am., 129.5: EL190-EL196, 2011.", | |
| "links": null | |
| }, | |
| "BIBREF31": { | |
| "ref_id": "b31", | |
| "title": "Towards end-to-end speech recognition with deep convolutional neural networks", | |
| "authors": [ | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Zhang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Pezeshki", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "P", | |
| "middle": [], | |
| "last": "Brakel", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Zhang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "C", | |
| "middle": [], | |
| "last": "Laurent", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Y", | |
| "middle": [], | |
| "last": "Bengio", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Courville", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2016, | |
| "venue": "Proc. of INTERSPEECH", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "410--414", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Y. Zhang, M. Pezeshki, P. Brakel, S. Zhang, C. Laurent, Y. Bengio, and A. Courville, \"Towards end-to-end speech recognition with deep convolutional neural networks.\" in Proc. of INTERSPEECH, pp. 410-414, 2016.", | |
| "links": null | |
| }, | |
| "BIBREF32": { | |
| "ref_id": "b32", | |
| "title": "Robust speech recognition from ratio masks", | |
| "authors": [ | |
| { | |
| "first": "Z.-Q", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2016, | |
| "venue": "Proc. of ICASSP", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "5720--5724", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Z.-Q. Wang and D. Wang., \"Robust speech recognition from ratio masks.\" in Proc. of ICASSP, pp. 5720- 5724, 2016.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "TABREF1": { | |
| "html": null, | |
| "num": null, | |
| "type_str": "table", | |
| "text": "\u5247\u4ee3\u8868 24 \u500b\u4e8c\u7dad\u7684 STMF \u7684\u8108\u885d\u97ff\u61c9\uff0c\u5176\uf96b\uf969\u5206\u5225\u70ba = { 4, 8, 16, 32 } Hz, \u03a9 = { 0.5, 1, 2 } cyc/otc\uff0c\u4e14\u5176\u65b9\u5411\u6027\u70ba\u96d9\u5411\u3002\u800c\u5716(2.6)\u5247\u662f\u67d0\uf9b5\uf906\u7d93\u904e\u521d\u671f\u8033\u8778\u968e\u6bb5\u5f8c\u6240\u5f97\u5230\u7684\u807d \u89ba\u983b\u8b5c\u518d\u7d93\u904e 8 \u7a2e\uf967\u540c\u7684 STMF \uf984\u6ce2\u5f8c\u4e4b\u7d50\u679c\u3002 \u5716 2.5\uff1aSTMFs \u4e8c\u7dad\u7684\u8108\u885d\u97ff\u61c9\u7bc4\uf9b5 \u5716 2.6\uff1a\u7d93\u904e\u521d\u671f\u8033\u8778\u968e\u6bb5\u5f8c\u6240\u5f97\u5230\u7684\u807d\u89ba\u983b\u8b5c\u518d\u7d93\u904e 8 \u7a2e\uf967\u540c\u7684 STMF \uf984\u6ce2\u5f8c\u4e4b\u7d50\u679c \u5be6\u9a57\u8a9e\uf9be \u672c\uf941\u6587\u4e2d\uff0c\u6211\u5011\u4f7f\u7528 2008 NIST SRE (Speaker Recognition Evaluation)\u8cc7\uf9be\u5eab\u4e2d\u7684\u8a9e\u97f3\u8a0a\u865f\uff0c\u6b64 \u8cc7\uf9be\u5eab\u662f\u7531\u8a9e\u8a00\uf969\u64da\uf997\u76df(Linguistic Data Consortium, LDC)\u53ca\u7f8e\u570b\u570b\u5bb6\u6a19\u6e96\u6280\u8853\u7814\u7a76\u6240(National Institute of Standards and Technology, NIST)\u6240\u63d0\u51fa\u3002 \u6211\u5011\u6240\u7528\u7684\u8cc7\uf9be\u70ba training set \u4e2d short2 \u7684\u96fb\u8a71\u8a9e\uf9be\uff0c\u6bcf\u500b\u97f3\u6a94\u7d04 5 \u5206\u9418\uff0c\u5de6\u53f3\u8072\u9053\u5206\u5225\u70ba \uf967\u540c\u7684\u8a9e\u8005\u3002\u6211\u5011\u96a8\u6a5f\u62bd\u53d6 100 \u4eba\uff0c\u4e26\u5c07\u975c\u97f3\u7684\u90e8\u5206\u5148\ufa08\u79fb\u9664\u3001\u5408\u4f75\uff0c\u518d\u5c07\u5176\ufa00\u6210 24 \u4efd 5 \u79d2 \u7684\u97f3\u6a94\uff0c\u4e26\u52a0\u5165\uf9ba\u7531 NOISEX-92 \u8cc7\uf9be\u5eab\u53d6\u5f97\u7684\u80cc\u666f\u96dc\u8a0a\uff0c\u8a0a\u96dc\u6bd4\u6703\u5728\u7b2c\u56db\u7ae0\u8a73\u8ff0\u5be6\u9a57\u7d50\u679c\u6642\u9032 \ufa08\uf96f\u660e\u3002\u800c\u70ba\uf9ba\u78ba\u4fdd\u6e2c\u8a66\u97f3\u6a94\u7684\u53ef\u4fe1\u53ca\u7a69\u5b9a\ufa01\uff0c\u6211\u5011\u5f9e 24 \u4efd\u97f3\u6a94\u4e4b\u4e2d\uff0c\u6311\u9078\u51fa 2 \u500b\u80fd\uf97e\u6700\u5f37\uff0c \u4e5f\u5c31\u662f\u8a9e\u97f3\u8cc7\u8a0a\u6700\u8c50\u5bcc\u7684\u97f3\u6a94\u7576\u4f5c\u6e2c\u8a66\u7528\u8cc7\uf9be\uff0c\u800c\u5269\u9918\u7684 22 \u4efd\u97f3\u6a94\u5247\u7576\u4f5c\u8a9e\u8005\u6a21\u578b\u4e4b\u8a13\uf996\u8cc7 \uf9be\u3002 \u8ef8\u5927\u5c0f\u70ba 15\uff0c\u6839\u64da\u4e00\u7dad\u5377\u7a4d\u6838 10ms \u70ba\u4e00\u9593\u683c\uff0c\u6211\u5011\u53ef\u4ee5\u5f97\u77e5\uff0cy \u8ef8\u5927\u5c0f\u70ba 15 \u7684\uf9fa \u6cc1\u4e0b\uff0c\u80fd\u5920\u5305\u542b 150ms \u7684\u8cc7\u8a0a\uff1b\u800c\u5728\u983b\uf961\u8ef8\u4e0a\uff0c\u6211\u5011\u9078\u64c7 x \u8ef8\u5927\u5c0f\u70ba 7 \u7684\u5377\u7a4d\u6838\uff0c\u662f\u56e0\u70ba\u9019 \u500b\u5927\u5c0f\u80fd\u5920\u5305\u542b\uf978\u500b\u516b\ufa01\u97f3\uff0c\u4ee5\u4eba\u5e73\u5e38\u5728\u8b1b\u8a71\u70ba\uf9b5\u5b50\uff0c\u5373\u53ef\u4ee5\u5305\u542b\u80fd\uf97e\u8f03\u70ba\u660e\u986f\u7684\u7b2c\u4e00\u5171\u632f\u5cf0\uff0c \u56e0\u6b64\u5728\u5377\u7a4d\u7684\u904e\u7a0b\u4e2d\uff0c\u6211\u5011\u53ef\u4ee5\u6709\u6548\u7684\u64f7\u53d6\u51fa\u8f03\u6709\u610f\u7fa9\u7684\u80fd\uf97e\u5340\u584a\u4e2d\u7684\u96b1\u85cf\u8cc7\u8a0a\u3002 \u5176\u4e2d\uff0c f (Hz) \u662f\u4e2d\u5fc3\u983b\uf961\uff0c\uf066 \u662f\u8f09\u6ce2\u76f8\u4f4d\uff0c a \u662f\u632f\u5e45\uff0c n \u662f\uf984\u6ce2\u5668\u7684\u9806\u5e8f\uff0c b (Hz) \u662f\uf984\u6ce2 \u7684\u5377\u7a4d\u6838\uff0c\u662f\u56e0\u70ba 7 \u80fd\u5920\u6db5\u84cb\uf978\u500b\u516b\ufa01\u97f3(octave)\uff0c\u5f9e\u5716(3.5) \u7c21\u55ae\uf92d\u770b\uff0c\u7de8\u865f\u7b2c 14 \u5230\u7de8\u865f\u7b2c 20 \u500b\u4e00\u7dad\u5377\u7a4d\u6838\u6240\u6db5\u84cb\u7684\u983b\uf961\u7bc4\u570d\uff0c\u5373\u662f 1000Hz~4000Hz\uff0c\u4e5f \u5c31\u662f\uf978\u500b\u516b\ufa01\u97f3\u3002 1x200 \u7684\u5377\u7a4d\u6838\uff0c\u4e26 \u4ee5\u6bcf 80 \u9ede\u505a\u4e00\u6b21\u5e73\u79fb\u76f8\u4e58\uff1b\u800c\u4e8c\u7dad\u5377\u7a4d\u5c64\u5305\u542b 24 \u500b 7x15 \u7684\u5377\u7a4d\u6838\uff1b\u5f8c\u9762\uf99a\u63a5\u5927\u5c0f\u70ba 1x5 \u7684 \u6700\u5927\u6c60\u5316\u5c64\uff0c\u4e26\u5728\u4e4b\u5f8c\u63a5\u4e0a 4 \u5c64\u7bc0\u9ede\uf969\u70ba 512 \u7684\u7279\u5fb5\u6574\u5408\u5c64\u3002\u5982\u5716(3.4)\u6240\u793a\u3002", | |
| "content": "<table><tr><td>\u5716 2.4\uff1a\u8a9e\u97f3\u8a0a\u865f\u7d93\u5728\u807d\u89ba\u6a21\u578b\u4e2d\u7d93\u904e\u521d\u671f\u8033\u8778\u968e\u6bb5\u6240\u7522\u751f\u4e4b\u807d\u89ba\u983b\u8b5c \u6700\u5f8c\uff0c\u8a0a\u865f\u6703\u901a\u904e\u4e00\u5c01\u5305\u64f7\u53d6\u5668\uff0c\u5982\u5f0f(2.4)\u3002\u800c\u5176\u7a4d\u5206\u7a97\u51fd\u5f0f\u5247\u5beb\u6210\u5f0f\u5b50(2.6)\uff1a ) ( ) ; ( t e t u t \uf06d \uf074 \uf074 \uf0b4 \uf03d \uf02d \u7d93\u904e\u4ee5\u4e0a\u7684\u8655\uf9e4\uff0c\u6211\u5011\u53ef\u4ee5\u5f97\u5230 \uff0c\u4e5f\u5c31\u662f\u6642\u983b\u57df\u7684\u807d\u89ba\u983b\u8b5c\u5716(auditory spectrogram)\u3002 (2.6) \u548c\u4e00\u822c\u7684\u77ed\u6642\u5085\uf9f7\uf96e\u8f49\u63db\u983b\u8b5c\u5716(STFT spectrogram)\uf967\u540c\u4e4b\u8655\u70ba\uff0c\u6b64\u983b\u8b5c\u7684\u983b\uf961\u8ef8\u662f\u4ee5\u5c0d\uf969\u5448\u73fe\uff0c \u800c\u983b\u5bec \u548c \u6703\u6839\u64da\u4e2d\u5fc3\u983b\uf961 \u548c\u03a9 \u800c\u589e\u52a0\u3002 \u7576\u900f\u904e\u5377\u7a4d\u5c64\u5f97\u5230\u7279\u5fb5\u5716\u4e4b\u5f8c\uff0c\u6211\u5011\u5e0c\u671b\u80fd\uf9dd\u7528\u9019\u4e9b\u7279\u5fb5\uf92d\u505a\u5206\uf9d0\uff0c\u4f46\u662f\u5c0d\u65bc\u4e00\u500b\u592a\u5927\u7684 \u7279\u5fb5\u8f38\u5165\u5206\uf9d0\u5668\uf92d\uf96f\uff0c\u9700\u8981\u904e\u65bc\u9f90\u5927\u7684\u8a08\u7b97\uf97e\u800c\u4e14\u5f88\u5bb9\uf9e0\u51fa\u73fe\u904e\u64ec\u5408(over-fitting)\u7684\u60c5\u5f62\u3002\u56e0 \u4fc2\uff0c\u9078\u64c7\u4f7f\u7528 20 \u500b\u5377\u7a4d\u6838\u9032\ufa08\uf984\u6ce2\uff0c\u4e26\u5c07\u6bcf\u500b\uf984\u6ce2\u5668\u6240\u5f97\u5230\u7684\u7d50\u679c\u9032\ufa08\u6392\uf99c\uff0c\u53ef\u4ee5\u5f97\u5230\u57fa\u65bc \u807d\u89ba\u611f\u77e5\u6a21\u578b\u521d\u671f\u8033\u8778\u968e\u6bb5\u7684\u807d\u89ba\u983b\u8b5c\u3002 \u5728\u807d\u89ba\u611f\u77e5\u6a21\u578b\u4e2d\uff0c\u8072\u97f3\u50b3\u81f3\u8033\u8778\u5f8c\uff0c\u6703\u5728\u57fa\u5e95\u819c\u4e0a\u4f9d\u7167\u5176\u672c\u8eab\uf967\u540c\u7684\u983b\uf961\uff0c\u800c\u88ab\uf967\u540c\u7684 \u807d\ufa19\u7d93\u5143\u89e3\u6790\u51fa\uf92d\u3002\u800c\u6211\u5011\u6240\u63d0\u51fa\u7684\u6a21\u578b\u4e2d\uff0c\u4e00\u7dad\u5377\u7a4d\u5c64\u5373\u662f\u8981\u6a21\u64ec\u8033\u8778\u5206\u983b\u7684\u52d5\u4f5c\uff0c\u4e5f\u5c31\u662f 4. \u5be6\u9a57\u8207\u8a0e\uf941 \u6838\u7684\u5f62\uf9fa\uff0c\u4e26\u6aa2\u8996\u6b64\u6a21\u578b\u67b6\u69cb\u4e0b\u8207\u524d\uf969\u7a2e\u6a21\u578b\u4e4b\u6548\ufa17\u6bd4\u8f03\u3002 \u4ee5\u53ca\u900f\u904e\u8a13\uf996\u53bb\u4fee\u6b63\u5176\u503c\uf978\uf9d0\uff0c\u5176\u7d50\u679c\u5982\u5716(4.1)\u6240\u793a\u3002\u56e0\u70ba\u56fa\u5b9a gammatone \uf984\u6ce2\u5668\u7684\u7d50\u679c\u4e26 \uf967\u6703\u56e0\u70ba\u662f\u5426\u7d66\u5b9a\u4e8c\u7dad\u5377\u7a4d\u6838\u521d\u59cb\u503c\u800c\u6709\u6240\u5dee\uf962\uff0c\u6545\u5716(4.1)\u5de6\u908a\uff0c\u4ee3\u8868\u8457\u6a21\u578b gammafix_A1init \u4ecd\u4fdd\uf9cd\u8457 gammatone \uf984\u6ce2\u5668\u7684\u983b\uf961\u9078\u64c7\u7279\u6027\uff0c\u56e0\u6b64\uff0c\u5728\u9019\u500b\u968e\u6bb5\u6211\u5011\u5c07\u91dd\u5c0d\u7d93\u904e\u4e00\u7dad\u5377\u7a4d\u6838 \u6240\u5f97\u5230\u7684\u807d\u89ba\u983b\u8b5c\u5716\uff0c\u7d93\u904e\u7b2c\u4e8c\u968e\u6bb5\uff0c\u4e5f\u5c31\u662f\u4eff\u5927\u8166\u76ae\u8cea\u968e\u6bb5\u7684\u4e8c\u7dad\u5377\u7a4d\u6838\u9032\ufa08\u8a0e\uf941\u3002 \u7528\uf92d\u8fa8\uf9fc\u8a9e\u8005\u3002\u6211\u5011\u900f\u904e\u7d66\u4e88\u5177\u6709\u5176\u7269\uf9e4\u610f\u7fa9\u7684\uf978\u968e\u6bb5\u5377\u7a4d\u5c64\u4e4b\u5377\u7a4d\u6838\u521d\u59cb\u503c\uff0c\u518d\uf9dd\u7528\uf9d0\ufa19\u7d93 \u7db2\uf937\u524d\u994b\u4ee5\u53ca\u53cd\u5411\u50b3\u64ad\u6f14\u7b97\u6cd5(feed-forward and back-propagation)\u9032\ufa08\u8a13\uf996\uff0c\u4e26\u6839\u64da\u8a9e\u8005\uf9fc\u5225\u7684 \u5716(2.5)3. \uf9d0\ufa19\u7d93\u7db2\uf937\u7cfb\u7d71\u67b6\u69cb\u8207\uf96b\uf969\u8a2d\u5b9a 3.1 \u5377\u7a4d\ufa19\u7d93\u7db2\uf937\u7c21\u4ecb \u6b64\u6211\u5011\u5e0c\u671b\u5f97\u5230\u7684\u7279\u5fb5\u5716\u5177\u6709\u5e73\u79fb\uf967\u8b8a\u6027\uff0c\u4e26\u900f\u904e\u9019\u500b\u7279\u6027\u5c07\u5c0d\u65bc\uf967\u540c\u4f4d\u7f6e\u7684\u7279\u5fb5\u503c\u9032\ufa08\u805a\u5408 \u7d71\u8a08\uff0c\u4e00\u822c\uf92d\uf96f\u5c31\u662f\u8a08\u7b97\u67d0\u500b\u7279\u5b9a\u5340\u57df\u7684\u6700\u5927\u503c\u6216\u5e73\u5747\u503c\uff0c\u800c\u9019\u7a2e\u805a\u5408\u7d71\u8a08\u7684\u904e\u7a0b\u5c31\u7a31\u70ba\u6c60\u5316 (pooling)\uff0c\u5716(3.3)\u986f\u793a\u4e00\u6700\u5927\u6c60\u5316\u7684\uf9b5\u5b50\u3002 \u5f97\u5230\u807d\u89ba\u983b\u8b5c\u5f8c\uff0c\u5728\u4e8c\u7dad\u7684\u5377\u7a4d\u5c64\u6642\uff0c\u6211\u5011\u9078\u7528\uf9ba 24 \u500b 7x15 \u7684\u5377\u7a4d\u6838\uff0c\uf92d\u6a21\u64ec\u5927\u8166\u76ae\u8cea \u968e\u6bb5\u6642\uff0c\u6703\u5c0d\u807d\u89ba\u983b\u8b5c\u4f5c\u4e00\u500b\u4e8c\u7dad\u8abf\u8b8a\u8cc7\u8a0a\u7684\u64f7\u53d6\u52d5\u4f5c\u3002\u800c\u6c60\u5316\u5c64\u5247\u662f\u5c07\u6211\u5011\u6240\u5f97\u5230\u7684\u7d50\u679c\uff0c \u4fdd\uf9cd\u91cd\u8981\u8cc7\u8a0a\u4e26\u9032\ufa08\ufa09\u7dad\uff0c\uf92d\ufa09\u4f4e\u6211\u5011\u6574\u9ad4\u7684\u904b\u7b97\uf97e\u3002\u800c\u7279\u5fb5\u6574\u5408\u5c64\u5247\u662f\u5c07\u6211\u5011\u6240\u5f97\u5230\u7684\u8cc7\u8a0a \u9032\ufa08\u7d71\u6574\u3001\u5206\u6790\uff0c\u85c9\u4ee5\u6a21\u64ec\u5927\u8166\uf901\u9ad8\u968e\u5c64\u7684\u8cc7\u8a0a\u7d71\u6574\u52d5\u4f5c\u3002 \uf9dd\u7528 20 \u7d44\u5e36\u901a\uf984\u6ce2\u5668\uff0c\u91dd\u5c0d\u5404\u500b\u4f4d\u7f6e\u7684\u5171\u632f\u97ff\u61c9\uff0c\uf92d\u5c0d\u539f\u59cb\u8a0a\u865f\u9032\ufa08\uf984\u6ce2\u3002\u56e0\u70ba\u57fa\u5e95\u819c\u5c0d\u65bc \uf967\u540c\u4f4d\u7f6e\u7684\u8072\u97f3\u97ff\u61c9\u904e\u7a0b\u76f8\u7576\u65bc\u4e00\u500b\uf984\u6ce2\u904e\u7a0b\uff0c\u800c\u73c8\u746a\u5f62\uf9fa\uf984\u6ce2\u5668(gammatone filter)\u7d50\u5408\uf9ba\u4eba \u8033\u7684\u807d\u89ba\u7279\u6027\uff0c\u4e5f\u5c31\u662f\u5c0d\u4e2d\u5fc3\u983b\uf961\u5448\u5c0d\uf969\u5206\u5e03\uf92d\u6a21\u64ec\u57fa\u5e95\u819c\u7684\u7279\u6027\uff0c\u5176\uf969\u5b78\u5f0f\u5982\u4e0b\uff1a 1 2 ( ) (2 ) n b t g t at e cos ft \uf070 \uf070 \uf066 \uf02d \uf02d \uf03d \uf02b (3.1) \u5728\u6642\u9593\u8ef8\u4e0a\uff0c\u56e0\u70ba 50ms \u5927\u7d04\u662f\u4eba\u80fd\u5920\uf9e4\u89e3\u8a9e\u97f3\u4e2d\u7684\u6700\u5c0f\u55ae\u4f4d\u7684\u6642\u9593\uff0c\u4f46\u6211\u5011\u53c8\u5e0c\u671b\u80fd\u5920 \u5206\u6790\u5230\u8f03\u5c0f\u7684 rate \u6240\u5305\u542b\u7684\u8a9e\u97f3\u9577\u6642\u8cc7\u8a0a\uff0c\u56e0\u6b64\u6211\u5011\u9078\u5b9a 150ms \u70ba\u6211\u5011\u5377\u7a4d\u6838 x \u8ef8\u7684\u5927\u5c0f\uff0c \u800c\u5176\u5012\uf969\uff0c\u4e5f\u5c31\u662f 6.7Hz\uff0c\u662f\u5206\u6790\u6240\u5f97\u5230\u7684\u807d\u89ba\u983b\u8b5c\u7684\u8a9e\u97f3\u5c01\u5305\u8b8a\u5316\u6700\u5c0f\u55ae\u4f4d\uff0c\u4f46\u5728\u5927\u5c0f\u70ba 150ms \u7684\u97f3\u6846\u4e0a\uff0c\u6211\u5011\u53ef\u4ee5\u770b\u5230\u6ce2\u9577\u70ba 250ms \u7684\u4e00\u534a\u4ee5\u4e0a\u7684\u6ce2\u578b\uff0c\u6545\u9019\u500b\u6642\u9593\u8ef8\u7684\u5927\u5c0f\uff0c\u53ef 4.1 \u6bd4\u8f03\u7cfb\u7d71\u4ecb\u7d39 \u6839\u64da\u807d\u89ba\u611f\u77e5\u6a21\u578b\u7684\u7279\u6027\uff0c\u6211\u5011\u5c07\u8003\u616e\u4e94\u5927\uf9d0\u6a21\u578b\uf92d\u9032\ufa08\u5be6\u9a57\uff0c\u5982\u4e0b\u8868(4.1)\u6240\u793a\uff0c\u4ee5\u4e0b\u91dd \u5c0d\u5404\uf9d0\u6a21\u578b\u7684\u8a2d\u5b9a\u505a\uf96f\u660e\uff1a Model 1D CNN kernel type 2D CNN kernel type Referred to \u53ca gammafix_A1rand \u6240\u56fa\u5b9a\u7684\u4e00\u7dad\u5377\u7a4d\u6838\u983b\uf961\u632f\u5e45\u97ff\u61c9\u3002 \u6211\u5011\u53ef\u4ee5\u5f9e\u5716(4.1)\u4e2d\u53f3\u908a\uf978\u5716\u767c\u73fe\uff0c\u80fd\u5920\u900f\u904e\u8a13\uf996\u800c\u4fee\u6b63\u4e00\u7dad\u5377\u7a4d\u6838\u7684\u6a21\u578b\uff0c\u5176\u5377\u7a4d\u6838\u5927 \u81f4\u4e0a\u4ecd\u4fdd\uf9cd\u8457 gammatone \uf984\u6ce2\u5668\u7684\u983b\uf961\u9078\u64c7\u7279\u6027\uff0c\u4f46\u662f\u5c0d\u65bc\uf967\u540c\u983b\u5e36\uff0c\u537b\u6709\u8457\uf967\u540c\u5f37\ufa01\u7684\u589e \ufa17\u3002\u4ee5\u9ad8\u983b\u7684\u5377\u7a4d\u6838\uf92d\uf96f\uff0c\u5176\u660e\u986f\u6bd4\u5de6\u5716\u4e2d\u539f\u59cb\u7684\u9ad8\u983b\u5377\u7a4d\u6838\uff0c\u80fd\uf97e\uf92d\u7684\u5f37\u3002\u56e0\u6b64\u6211\u5011\u53ef\u4ee5\u63a8 \u5f9e\u8868(4.2)\u4e2d\u6211\u5011\u53ef\u4ee5\u767c\u73fe\uff0c\u5728\u6709\u7d66\u5b9a gammatone \uf984\u6ce2\u5668\u7d50\u679c\u4e4b\u524d\u56db\u7a2e\u6a21\u578b\uff0c\ufa26\u6709\u8457\u5dee\uf967 \u591a\u7684\u8868\u73fe\uff0c\u5373\u4f7f\u56fa\u5b9a gammatone \uf984\u6ce2\u5668\u7684\u6a21\u578b\uff0c\u5728 -5dB \u8a0a\u96dc\u6bd4\u7684\uf9fa\u6cc1\u4e0b\u6709\u8457\u7a0d\u5fae\u8f03\u5dee\u7684\u8868\u73fe\uff0c \u4f46\u5176\u5728 0dB \u53ca 5dB \u4e0a\u4ecd\u6709\u76f8\u4f3c\u7684\u8868\u73fe\u3002\u56e0\u6b64\u6211\u5011\u63a8\uf941\uff0c\u9019\u56db\u7a2e\u6a21\u578b\ufa26\u6709\u8457\uf9d0\u4f3c\u529f\u80fd\u7684\u4e8c\u7dad\u5377 \u7a4d\u6838\u3002\u800c\u5176\u7d50\u679c\u5982\u5716(4.2)\u6240\u793a\u3002 \u76ee\u6a19\uf92d\u6539\u5584\u6a21\u578b\u7684\u6027\u80fd\u8868\u73fe\u3002\u800c\u6211\u5011\u4e5f\u53ef\u4ee5\u900f\u904e\u7d93\u904e\u8a13\uf996\u5f8c\u800c\u8abf\u6574\u7684\u5377\u7a4d\u6838\u767c\u73fe\uff0c\u7121\uf941\u662f\u5728\u7b2c \u4e00\u968e\u6bb5\u7684\u8033\u8778\u5206\u983b\u4ea6\u6216\u662f\u7b2c\u4e8c\u968e\u6bb5\u7684\u5927\u8166\u76ae\u8cea\u968e\u6bb5\uff0c\u6211\u5011\u7686\u53ef\u4ee5\u900f\u904e\u5176\u8a13\uf996\u8abf\u6574\u5f8c\u7684\u5377\u7a4d\u6838\u5f62 \uf9fa\uff0c\u9032\ufa08\u5224\uf95a\u8207\u5206\u6790\u3002\u9019\u7a2e\u900f\u904e\u8f38\u5165\u539f\u59cb\u8a0a\u865f(raw data)\u7684\u67b6\u69cb\uf9e4\uf9a3\uff0c\u4e5f\u8a31\u53ef\u4ee5\u548c\u4ee5\u807d\u89ba\u79d1\u5b78 \u4f5c\u70ba\u57fa\u790e\u7684\uf96b\uf969\u7cfb\u7d71\u505a\u6bd4\u8f03\uff0c\u540c\u6a23\u7684\uff0c\u6211\u5011\u4e5f\u53ef\u4ee5\u900f\u904e\u521d\u59cb\u5316\u5377\u7a4d\u6838\uff0c\uf92d\u4f7f\u6a21\u578b\u5728\u76f8\u540c\u7684\u6642\u9593 \u689d\u4ef6\u4e2d\u6216\u8005\u8f03\u5c11\u7684\u8cc7\uf9be\uf97e\u4e0b\uff0c\u5176\u8868\u73fe\u512a\u65bc\uf967\u7d66\u4e88\u4efb\u4f55\u521d\u59cb\u503c\u7684\u6a21\u578b\uff0c\u9019\u4ee3\u8868\u8457\u5373\u4f7f\u5728\u8f03\u70ba\u56b4\u82db \u9019\uf9e8\u6240\u6709\uf96b\u8207\u6bd4\u8f03\u7684\u6a21\u578b\u5747\u6709\u540c\u6a23\u7684\u67b6\u69cb\uff0c\u4ea6\u5373\u4e00\u7dad\u5377\u7a4d\u5c64\u5305\u542b 20 \u500b 4.2 \u5be6\u9a57\u7d50\u679c \uf941\uff0c\u6a21\u578b\u7d93\u904e\u8a13\uf996\u5f8c\uff0c\u7684\u78ba\u6703\u6839\u64da\u61c9\u7528\u76ee\u7684\u7684\uf967\u540c\uff0c\u6216\u8005\u80cc\u666f\u96dc\u8a0a\u7684\uf967\u540c\uff0c\uf92d\u8abf\u6574\u5728\u8a72\u76ee\u7684\u4e4b \u60c5\u6cc1\u4e0b\uff0c\u6211\u5011\u4e5f\u53ef\u4ee5\u900f\u904e\u7d66\u4e88\u5377\u7a4d\u6838\u521d\u59cb\u503c\uff0c\u4f7f\u5176\u671d\u8457\u9019\u500b\u65b9\u5411\u9032\ufa08\u5fae\u8abf\u4fee\u6b63\uff0c\uf92d\u9054\u5230\u8f03\u597d\u7684</td></tr><tr><td>x \u5373\u70ba\u57fa\u5e95\u819c\u4e0a \u8ddd\uf9ea\u8033\u8778\u5e95\u90e8\u7684\u8ddd\uf9ea\uff0c\u800c\u6a21\u578b\u4e2d\u4f7f\u7528 128 \u500b\u5177\uf967\u540c\u4e2d\u5fc3\u983b\uf961\u53ca\u983b\u5bec\u7684\u5e36\u901a\uf984\u6ce2\u5668\u7d44(band pass filter bank)\uf92d\u6a21\u64ec\u5404\u4f4d\u7f6e\u7684\u5171\u632f\u97ff\u61c9\uff0c\u5176\u4e2d\u4e2d\u5fc3\u983b\uf961\u548c\u983b\u5bec\u6210\u5e38\uf969 Q(constant Q)\u7684\u95dc\u4fc2\uff0c\u5982\u5f0f (2.5) \u4e2d\u5fc3\u983b\uf961 \u983b\u5bec \u5e38\uf969 (2.5) \u4e2d\u5fc3\u983b\uf961\u5728\u5c0d\uf969\u8ef8\u4e0a\u662f\u5747\u52fb\u5206\u5e03\u7684\uff0c\u63a5\u8457\uff0c\u6bcf\u500b\uf984\u6ce2\u5668\u7684\u8f38\u51fa\u5c07\u88ab\u50b3\u9001\u5230\u4e00\u975e\u7dda\u6027\u58d3\u7e2e\u968e \u6bb5\uff0c\u5c0d\u61c9\u5230\u5f0f(2.2)\u3002\u9019\u500b\u975e\u7dda\u6027\u58d3\u7e2e\u662f\u7528\uf92d\u6a21\u64ec\u8033\u8778\u57fa\u5e95\u819c\u7684\u9707\u52d5\u8f49\u5316\u6210\u5167\u6bdb\u7d30\u80de\u7684\u96fb\u4f4d\uff0c\u800c \u5982\u5716(2.4)\u6240\u793a\uff0c\u63a5\u8457\u5728\u7b2c\u4e8c\u968e\u6bb5\u7684\u5927\u8166\u76ae\u8cea\u5206\u6790\u5c07\u91dd\u5c0d\u6b64\u807d\u89ba\u983b\u8b5c\u4f5c\u9032\u4e00\u6b65\u7684\u5206\u6790\u3002 2.4 \u807d\u89ba\u611f\u77e5\u6a21\u578b-\u5927\u8166\u76ae\u8cea\u968e\u6bb5 \u7b2c\u4e8c\u968e\u6bb5\u662f\u5728\u6a21\u64ec\u5927\u8166\u76ae\u8cea A1 \u5340\u7684\ufa19\u7d93\u5143\u5c0d\u65bc\u6642\u983b\u7684\u9078\u64c7\u6027\u3002\u5728\u807d\u89ba\u6a21\u578b\u4e2d\uff0c \u662f\u7d93 \u904e\u521d\u671f\u8033\u8778\u968e\u6bb5\u6240\u5f97\u5230\u7684\u807d\u89ba\u983b\u8b5c\u5716\u3002\u5927\u8166\u76ae\u8cea A1 \u5340\u7684\ufa19\u7d93\u53ef\u4ee5\u88ab\u8996\u70ba\u4e00\u7cfb\uf99c\u5177\u6709\uf967\u540c\u7279\u5fb5 \uf96b\uf969\u7684\u4e8c\u7dad\u6642\u983b\u8abf\u8b8a\uf984\u6ce2\u5668 (spectro-temporal modulation filters , STMFs)\uff0c\u53ef\u4ee5\u7528\uf92d\u89e3\u6790\u6240\u5f97\u5230 \u7684\u807d\u89ba\u983b\u8b5c\u3002\u63db\uf906\u8a71\uf96f\uff0cA1 \u6a21\u578b\u5c07\u539f\u672c\u7684\u807d\u89ba\u983b\u8b5c\u6839\u64da\uf967\u540c\u7684\u6642\u983b\u8abf\u8b8a\u9032\ufa08\u89e3\u6790\uff0c\u6211\u5011\u5047\u8a2d\uff0c \u5728 A1 \u5f8c\u7684\ufa19\u7d93\u5143\u53ef\u4ee5\u6536\u96c6\u4e26\u6574\u5408\u8a31\u591a\u7d93 STMF \u89e3\u6790\u5f8c\u7684\u5177\u9ad4\u8cc7\u8a0a\uff0c\u9032\u800c\u5efa\u69cb\u51fa\uf901\u9ad8\u968e\u7684\u5927\u8166 \u8a8d\u77e5\u529f\u80fd\u3002 \u751f\u6210 STMF \u7684\uf96b\uf969\u5305\u542b\uf9ba rate (Hz) \u3001scale \u03a9 (cycle/octave) \u4ee5\u53ca\u65b9\u5411\u6027\u3002rate \u6355\u6349\uf9ba \u807d\u89ba\u983b\u8b5c\u6cbf\u8457\u6642\u9593\u8ef8\u7684\u8b8a\u5316\u901f\ufa01\uff0c\u800c scale \u5247\u662f\u6355\u6349\uf9ba\u5176\u6cbf\u8457\u983b\uf961\u8ef8\u7684\u80fd\uf97e\u5206\u5e03\uf9fa\u6cc1\uff0c\u6b64\u5916\uff0c rate \u7684\u7b26\u865f\u4ee3\u8868\uf9ba STMF \u7684\u65b9\u5411\u6027( \u6b63/\u8ca0 \u7b26\u865f\u4ee3\u8868 \u5411\u4e0b/\u5411\u4e0a \u7684\u65b9\u5411)\uff0c\u800c STMF \u7684\u983b\uf961\u97ff\u61c9 \u53ef\u4ee5\u5beb\u6210 (2.7)\u53ca(2.8)\uff1a \u5716 3.3\uff1a\u5927\u5c0f\u70ba 2x2 \u4e4b\u6700\u5927\u6c60\u5316\u7bc4\uf9b5 \u800c\u7279\u5fb5\u6574\u5408\u5c64\uff0c\u70ba\u5377\u7a4d\ufa19\u7d93\u7db2\uf937\u6700\u5f8c\u4e00\u500b\u968e\u6bb5\uff0c\u6b64\u5c64\u7684\u904b\u7b97\u65b9\u6cd5\u548c\u50b3\u7d71\ufa19\u7d93\u7db2\uf937\u76f8\u540c\uff0c\u5373 \u900f\u904e\u8f38\u5165\ufa19\u7d93\u5143\u548c\u8f38\u51fa\ufa19\u7d93\u5143\u9593\u4e92\u76f8\uf99a\u7d50\u800c\u6210\uff0c\u53ef\u628a\u524d\u9762\u63d0\u53d6\u4e4b\uf96b\uf969\u7528\u65bc\u5206\uf9d0(classification) \u6216\u56de\u6b78(regression)\u7684\u8b70\u984c\u4e0a\u3002 3.2 \u6a21\u578b\u67b6\u69cb \u5668\u7684\u983b\u5bec\uff0c t\u662f\u6642\u9593\u3002\u9019\u662f\u4e00\u500b\u4ee5\u73c8\u746a\u5206\u5e03(gamma distribution) \u51fd\uf969\uf92d\u8abf\u8b8a\u4e00\u55ae\u97f3\u7684\u51fd\u5f0f\u3002 \u56e0\u6b64\uff0c\u5728\u4e00\u7dad\u7684\u5377\u7a4d\u5c64\u6642\uff0c\u6211\u5011\u5e0c\u671b\u6240\uf984\u51fa\uf92d\u7684\u6ce2\u578b\u7684\u4e2d\u5fc3\u983b\uf961\uff0c\u80fd\u6839\u64da\u5176\u983b\u5bec\u6210\u5e38\uf969 Q (constant Q)\u95dc\u4fc2\uff0c\u6545\u6211\u5011\uf9dd\u7528\u73c8\u746a\u5f62\uf9fa\uf984\u6ce2\u5668\uf92d\u7522\u751f\u5177\u5e0c\u671b\u4e4b\u983b\uf961\u97ff\u61c9\u4e4b\uf984\u6ce2\u5668\u7d44\u3002\u4e0b\u5716(3.5) \u70ba\u6839\u64da\u5be6\u9a57\u8a2d\u5b9a\u6240\u5f97\u5230\u7684 20 \u500b\uf984\u6ce2\u5668\uff0c\u518d\u5206\u5225\u7d93\u904e 400 \u9ede\u7684\u5085\uf9f7\uf96e\u8f49\u63db\u6240\u5f97\u5230\u7684\u983b\uf961\u632f\u5e45\u97ff \u61c9\uff0c\u4e26\u4f9d\u7167\u5176\u4e2d\u5fc3\u983b\uf961\u4e4b\u9ad8\u4f4e\u6392\uf99c(x \u8ef8\uff0cfilter index)\u5f8c\u7684\u7d50\u679c\u3002 \u5927\u7d04\u5206\u6790 rate \u6700\u4f4e\u81f3 4Hz \u7684\u8a9e\u97f3\u5c01\u5305\u8b8a\u5316\u60c5\u5f62\u3002 \u518d\u52a0\u4e0a\u4e00\u7dad\u5377\u7a4d\u6838\u662f\u4ee5\u6bcf 10ms(\u4e5f\u5c31\u662f 100Hz)\u70ba\u97f3\u6846\u5f7c\u6b64\u9593\u7684\u9593\u9694\uff0c\u4ee5\u53d6\u6a23\u5b9a\uf9e4\u6211\u5011\u53ef\u4ee5 \u5f97\u77e5\uff0c\u6700\u9ad8\u89c0\u5bdf 50Hz \u7684\u8b8a\u5316\uf97e\u3002\u7d9c\u5408\u4ee5\u4e0a\uf978\u9ede\uff0c\u5728\u6642\u9593\u8ef8\u7684\u5206\u6790\u4e0a\uff0c\u6211\u5011\u53ef\u4ee5\u89c0\u5bdf\u5230 rate \u70ba 4~50Hz \u7684\u6642\u57df\u8abf\u8b8a\u8b8a\u5316\u60c5\u5f62\u3002 Gammatone Fix A1 initial Gammafix_A1init A1 random Gammafix_A1rand Gammatone Initial A1 initial Gammainit_A1init A1 random Gammainit_A1rand Both random Bothrand Gammafix_A1init\uff1a\u4e00\u7dad\u5377\u7a4d\u6838\u56fa\u5b9a\u70ba 20 \u500b gammatone \uf984\u6ce2\u5668\u4e4b\u7d50\u679c\uff0c\u8a13\uf996\u6642\u7121\u6cd5\u5c0d\u6b64\u4e00\u7dad \u5377\u7a4d\u6838\u7d44\u9032\ufa08\u4fee\u6b63\uff1b\u800c\u4e8c\u7dad\u5377\u7a4d\u6838\u7684\u521d\u59cb\u5f62\uf9fa\u7d66\u5b9a\u8a08\u7b97\u51fa\uf92d\u4e4b 24 \u500b STMF\uff0c\u5f8c\uf92d\u900f\u904e\u524d\u994b\u4ee5\u53ca \u53cd\u5411\u50b3\u64ad\u6f14\u7b97\u6cd5(feed-forward and back-propagation)\u9032\ufa08\u8a13\uf996\u3002\u9019\u500b\u6a21\u578b\u7684\u5047\u8a2d\u662f\u8033\u8778\u968e\u6bb5\u7684\u5206 \u983b\uff0c\u662f\u6c92\u6709\u8fa6\u6cd5\u4f9d\u7167\u61c9\u7528\u76ee\u7684\u7684\uf967\u540c\u800c\u9032\ufa08\u8abf\u6574\u7684\uff1b\u800c\u5927\u8166\u76ae\u8cea A1 \u5340\u53ef\u4ee5\u6839\u64da\u61c9\u7528\u76ee\u7684\u9032\ufa08 \u8abf\u6574\u3002\u6b64\u8a2d\u5b9a\u8207\u52d5\u7269\ufa19\u7d93\u5be6\u9a57\u6240\u89c0\u5bdf\u5230\u7684\u73fe\u8c61\uf9d0\u4f3c\u3002 Gammafix_A1rand\uff1a\u4e00\u7dad\u5377\u7a4d\u6838\u56fa\u5b9a\u70ba 20 \u500b gammatone \uf984\u6ce2\u5668\u4e4b\u7d50\u679c\uff0c\u8a13\uf996\u6642\u7121\u6cd5\u5c0d\u6b64\u4e00\u7dad \u5377\u7a4d\u6838\u7d44\u9032\ufa08\u4fee\u6b63\uff1b\u800c\u4e8c\u7dad\u5377\u7a4d\u6838\uf967\u7d66\u7279\u5b9a\u7684\u521d\u59cb\u503c\uff0c\u76f4\u63a5\u900f\u904e\u524d\u994b\u4ee5\u53ca\u53cd\u5411\u50b3\u64ad\u6f14\u7b97\u6cd5\u9032\ufa08 \u6211\u5011\uf9dd\u7528\u6240\u63d0\u51fa\u7684\uf9d0\ufa19\u7d93\u7db2\uf937\u6a21\u578b\uff0c\uf92d\u6a21\u64ec\u807d\u89ba\u6a21\u578b\u4e2d\uff0c\u521d\u671f\u8033\u8778\u968e\u6bb5\u5c0d\u65bc\u8072\u97f3\u8a0a\u865f\u7684\u5206 \u983b\uff1b\u4ee5\u53ca\u5927\u8166\u76ae\u8cea A1 \u5340\u5c0d\u65bc\u807d\u89ba\u983b\u8b5c\u7684\u6642\u983b\u9078\u64c7\u6027\u3002\u56e0\u6b64\u5728\u9019\u500b\u7ae0\u7bc0\u4e2d\uff0c\u6211\u5011\u9664\uf9ba\u5c07\u6bd4\u8f03\u4e94 \u7a2e\u6a21\u578b\u5c0d\u6b63\u78ba\uf961\u7684\u5f71\u97ff\uff0c\u540c\u6642\u4e5f\u6703\u91dd\u5c0d\u6211\u5011\u6240\u63d0\u51fa\u7684\uf9d0\ufa19\u7d93\u7db2\uf937\u6a21\u578b\u7d93\u904e\u8a13\uf996\u5f8c\uff0c\u8207\u50b3\u7d71\u7684\u807d \u89ba\u611f\u77e5\u6a21\u578b\u7684\u76f8\u95dc\u6027\u53ca\u610f\u7fa9\u9032\ufa08\u8a0e\uf941\u3002 \u5728\u9019\u500b\u5be6\u9a57\uf9e8\u6211\u5011\u5c07\uf978\u7a2e\uf967\u540c\u7684\u80cc\u666f\u96dc\u8a0a\u5206\u5225\u4ee5\u8a0a\u96dc\u6bd4-5\u30010\u30015dB \u8207\u8a9e\u97f3\u76f8\u6df7\uff0c\u4e00\u5171\u7522\u751f buccaneer \u53ca factory\u3002 \u4e0b\u8868(4.2)\u70ba\u6b64\u6b21\u5be6\u9a57\u7684\u5be6\u9a57\u7d50\u679c\uff0c\u5f9e\u4e2d\u53ef\u4ee5\u767c\u73fe\uff0c\u5728\u6709\uf96b\u8003 gammatone \uf984\u6ce2\u5668\u7684\u6a21\u578b\uff0c\u7121\uf941 \u662f\u5426\u56fa\u5b9a\u5176\u4e00\u7dad\u7684\u5377\u7a4d\u6838\uff0c\u5728\u8a9e\u8005\uf9fc\u5225\u4e0a\u7684\u6821\u80fd\ufa26\u6703\u6bd4\u4e00\u7dad\u3001\u4e8c\u7dad\ufa26\u96a8\u6a5f\u7d66\u5b9a\u521d\u59cb\u503c\u7684\u6a21\u578b\uf92d \u7684\u597d\u3002\u5728\u6b64\u6211\u5011\u5c07\u91dd\u5c0d\u4ee5\u4e0b\u5e7e\u9ede\u9032\ufa08\u8a0e\uf941\uff1a I \u524d\u56db\u7a2e\u6a21\u578b\uff0c\u5c0d\u4e00\u7dad\u5377\u7a4d\u6838\u7684\u5f62\uf9fa\u9032\ufa08\u8a0e\uf941\u3002 II \u524d\u56db\u7a2e\u6a21\u578b\uff0c\u5c0d\u4e8c\u7dad\u5377\u7a4d\u6838\u5f62\uf9fa\u9032\ufa08\u8a0e\uf941\u3002 \u4e0b\u91cd\u8981\u983b\u5e36\u8cc7\u8a0a\u7684\u6b0a\u91cd\u3002 \u800c\u6211\u5011\u4e5f\u53ef\u4ee5\u900f\u904e\u8868(4.2)\u7684\u7d50\u679c\u767c\u73fe\uff0c\u5728\u4f4e\u8a0a\u96dc\u6bd4\u4e4b\u4e0b\uff0c\u56fa\u5b9a gammatone \uf984\u6ce2\u5668\u7684\u6a21\u578b \u660e\u986f\u8868\u73fe\u8f03\u5dee\uff0c\u6545\u6211\u5011\u53ef\u4ee5\u5408\uf9e4\u7684\u63a8\uf941\uff0c\u56e0\u70ba\u5728\u4f4e\u8a0a\u96dc\u6bd4\u4e0b\uff0c\u539f\u59cb\u8a0a\u865f\u88ab\u7834\u58de\u7684\u8f03\u70ba\u56b4\u91cd\uff0c\u56e0 \u6b64\u9700\u8981\u6bd4\u8f03\u80fd\u5920\u51f8\u986f\u67d0\u4e9b\u8f03\uf967\u53d7\u566a\u97f3\u5f71\u97ff\u7684\u7279\u5b9a\u983b\u5e36\u7684\uf984\u6ce2\u5668\uff0c\u800c\u900f\u904e\u91cd\u8981\uf984\u6ce2\u5668\u6240\u5f97\u7684\u807d\u89ba \u983b\u8b5c\u5716\uff0c\u5728\u6a21\u578b\u5f8c\u9762\u7684\u968e\u6bb5\uff0c\u4e5f\u5c31\u662f\u5927\u8166\u76ae\u8cea\u968e\u6bb5\u64f7\u53d6\u8a9e\u97f3\u91cd\u8981\u8cc7\u8a0a\u6642\uff0c\u4e5f\u80fd\u8f03\u6709\u5e6b\u52a9\u3002 \u5716 4.2\uff1a\u591a\u8a0a\u96dc\u6bd4\u53ca\u591a\u96dc\u8a0a\u7a2e\uf9d0\u689d\u4ef6\u4e0b\uff0c\u5404\u6a21\u578b\u8a13\uf996\u5f8c\u4e4b\u4e8c\u7dad\u5377\u7a4d\u6838\u5f62\uf9fa\u7d50\u679c\u5716 \u5716 4.3\uff1a\u591a\u8a0a\u96dc\u6bd4\u53ca\u591a\u96dc\u8a0a\u7a2e\uf9d0\u689d\u4ef6\u4e0b\uff0c\u5404\u6a21\u578b\u8a13\uf996\u5f8c\u529f\u80fd\u76f8\u540c\u4e4b\u4e8c\u7dad\u5377\u7a4d\u6838\u5f62\uf9fa\u7d50\u679c\u5716 \u800c\u6211\u5011\u5c07\u91dd\u5c0d\u7279\u5b9a\u5377\u7a4d\u6838\u9032\ufa08\u8a0e\uf941\uff1a\u4ee5\u4e0b\u5716(4.4)\u70ba\uf9b5\uff0c\u5716\u4e2d\u6211\u5011\u53ef\u4ee5\u770b\u5230\u8a72\u5377\u7a4d\u6838\u7d04\u5305\u542b 0.6 \u500b\u6ce2\u9577\uff0c\u800c\u6211\u5011\u7684\u5377\u7a4d\u6838\u8a2d\u5b9a\u70ba\u53ef\u4ee5\u6db5\u84cb 150ms \u7684\u8cc7\u8a0a\uff0c\u56e0\u6b64\u900f\u904e\u8a08\u7b97\uff0c\u6211\u5011\u53ef\u4ee5\u5f97\u5230\u5176 \u6ce2\u578b\u70ba\u6ce2\u9577 250ms \u4e5f\u5c31\u662f\u983b\uf961\u8b8a\u5316\u70ba 4Hz \u7684\u5377\u7a4d\u6838\u3002 \u5716 4.5\uff1a\u7531\u5de6\u81f3\u53f3\u70ba\u64f7\u53d6\u8abf\u8b8a\u983b\uf961\u8b8a\u5316\u70ba 8\u300116\u300132Hz \u7684\u5377\u7a4d\u6838 \u7136\u800c\uff0c\u9664\uf9ba\u5f9e\u8abf\u8b8a\u983b\uf961\u8b8a\u5316\u4e0a\uf92d\u89c0\u5bdf\u8a13\uf996\u5f8c\u5f97\u5230\u7684\u5377\u7a4d\u6838\u4e4b\u5916\uff0c\u6211\u5011\u4e5f\u53ef\u4ee5\u767c\u73fe\u6709\u4e9b\u5377\u7a4d \u6838\u7684\u5340\u57df\u80fd\uf97e\u7279\u5225\u7684\u5f37\uff0c\u5982\u4e0b\u5716(4.6)\u6240\u793a\uff0c\u9019\u8868\u793a\u6b64\u5377\u7a4d\u6838\u9664\uf9ba\u5305\u542b 3.5 \u500b\u6ce2\u578b\uff0c\u4e5f\u5c31\u662f\u4ee3\u8868 \u64f7\u53d6\u8abf\u8b8a\u983b\uf961\u8b8a\u5316 23.3Hz \u7684\u8a9e\u97f3\u8cc7\u8a0a\u5916\uff0c\u5176\u80fd\uf97e\u5448\u73fe\u7684\u65b9\u5f0f\u5247\u662f\u4ee3\u8868\u8457\u64f7\u53d6\u8a9e\u97f3\u8cc7\u8a0a\u80fd\uf97e\u8f03 \u5927\u7684\u5171\u632f\u5cf0\u90e8\u5206\u3002 \u5716 4.6\uff1a\u64f7\u53d6\u8abf\u8b8a\u983b\uf961\u8b8a\u5316\u70ba 23.3Hz \u4ee5\u53ca\u8a9e\u97f3\u5171\u632f\u5cf0\u7684\u5377\u7a4d\u6838 \u5716 4.7\uff1a Bothrand \u6a21\u578b\u7684\u4e00\u7dad\u5377\u7a4d\u6838\u983b\uf961\u632f\u5e45\u97ff\u61c9\u5716 \u4e26\u4e14\u56e0\u70ba\u5176\u4e00\u7dad\u5377\u7a4d\u6838\u4e26\u6c92\u6709\u4f9d\u7167\u4e2d\u5fc3\u983b\uf961\u9ad8\u4f4e\u9806\u5e8f\u800c\u6392\uf99c\uff0c\u6545\u6240\u5f97\u5230\u7684\u7d50\u679c\u4e26\u975e\u6211\u5011\u6240 \uf9e4\u89e3\u7684\u807d\u89ba\u983b\u8b5c\u5716\uff0c\u56e0\u6b64\u5f9e\u5716(4.8)\u4e2d\u6211\u5011\u53ef\u4ee5\u770b\u5230\uff0cBothrand \u6240\u7522\u751f\u7684\u4e8c\u7dad\u5377\u7a4d\u6838\u5167\u5bb9\u986f\u5f97\u8f03 \u70ba\u96dc\uf91b\uff0c\u6240\u4ee5\u8981\u5f9e\u4e2d\u5224\uf95a\u51fa\u4efb\u4f55\u6709\u95dc\u8a9e\u97f3\u610f\u7fa9\u7684\u8cc7\u8a0a\u662f\u975e\u5e38\u56f0\u96e3\u7684\u3002 \u6536\u6582\u7d50\u679c\u3002 \u4eba\uf9d0\u7684\u807d\u89ba\u611f\u77e5\u7cfb\u7d71\uff0c\u4e26\u975e\u53ea\u7528\u65bc\u55ae\u4e00\u4e00\u7a2e\u76ee\u6a19\uff0c\u800c\u8fd1\uf98e\uf92d\uff0c\u6709\u8a31\u591a\u900f\u904e\u5377\u7a4d\ufa19\u7d93\u7db2\uf937 (CNN)\u6210\u529f\u5730\u61c9\u7528\u65bc\u81ea\u52d5\u8a9e\u97f3\uf9fc\u5225(automatic speech recognition , ASR) [4][29][30]\u7b49\u7b49\u8b70\u984c\u4e0a\u7684 \uf9b5\u5b50\uff0c \u56e0\u6b64\u6211\u5011\u5e0c\u671b\uff0c\u672a\uf92d\u80fd\u767c\u5c55\u4e00\u5957\u57fa\u65bc\u611f\u77e5\u807d\u89ba\u6a21\u578b\u4e26\u4e14\u540c\u6642\u61c9\u7528\u65bc\u591a\u7a2e\u76ee\u6a19\u7684\u67b6\u69cb\uff0c \uf9b5\u5982\uff1a\u540c\u6642\u61c9\u7528\u65bc\u8a9e\u97f3\u8fa8\uf9fc\u53ca\u8a9e\u97f3\u589e\u5f37\u3002\u800c\u5728\u6b64\u67b6\u69cb\u5e95\u4e0b\uff0c\u8a72\u6a21\u578b\u80fd\u5920\u96a8\u8457\u76ee\u6a19\u7684\u6539\u8b8a\u9032\ufa08\u672c 3.3 3.4 \u8a9e\u97f3\u8655\uf9e4\u80cc\u666f\u77e5\uf9fc\u8207\uf96b\uf969\u8a2d\u5b9a \u8868 4.1\uff1a\u4e94\u5927\uf9d0\u6bd4\u8f03\u4e4b\u6a21\u578b \uf9d1\u7a2e\uf967\u540c\u60c5\u5883\u4e0b\u7684\u8a9e\uf906\u540c\u6642\u5c0d\u6a21\u578b\u9032\ufa08\u8a13\uf996\u3002\u6211\u5011\u5728\u9019\u6b21\u7684\u5be6\u9a57\u4e2d\u9078\u5b9a\uf978\u7a2e\u80cc\u666f\u96dc\u8a0a\uff0c\u5206\u5225\u70ba \u8eab\uf96b\uf969\u7684\u5fae\u8abf\uff0c\uf92d\u9054\u5230\u76f8\u5c0d\u65bc\u5176\u61c9\u7528\u4e4b\u8f03\u597d\u7684\uf9fa\u614b\u3002</td></tr><tr><td>\u5167\u6bdb\u7d30\u80de\u7684\u98fd\u548c\u73fe\u8c61\u3002\u63a5\u8457\u76f8\u8fd1\u7684\u5167\u6bdb\u7d30\u80de\u5f7c\u6b64\u4e4b\u9593\u6703\u6709\u4e00\u968e\u5074\u6291\u5236\u4f5c\u7528(lateral inhibitory network ,LIN)\uff0c\u5982\u5f0f(3.3)\uff0c\u4e5f\u6a21\u64ec\uf9ba\u807d\u89ba\u4e0a\u9130\u8fd1\u983b\uf961\u7684\u906e\u853d\u6548\u61c9\u3002 , \u672c\uf941\u6587\u4e2d\u6211\u5011\u6a21\u578b\u7684\u8f38\u5165\u70ba 275ms \u7684\u7247\u6bb5\u8a9e\u97f3\uff0c\u6240\u6709\u97f3\u6a94\u7684\u53d6\u6a23\u983b\uf961\u5b9a\u5728 8k Hz\uff0c\u6b64\u8a2d\u5b9a \u5716 3.6\uff1a24 \u500b\u6839\u64da\uf967\u540c\u7684 rate-scale \uf96b\uf969\u6240\u5708\u51fa\uf92d\u7684\u4e8c\u7dad\u5377\u7a4d\u6838\u521d\u59cb\u503c III \u7b2c\u4e94\u7a2e\u6a21\u578b Bothrand \u7684\u7d50\u679c\u8a0e\uf941\u3002 \u8a13\uf996\u3002\u9019\u500b\u6a21\u578b\u7684\u5047\u8a2d\u8207\u7b2c\u4e00\uf9d0\u578b(Gammafix_A1init)\u76f8\u4f3c\uff0c\uf967\u540c\u7684\u5730\u65b9\u662f\u4e8c\u7dad\u5377\u7a4d\u6838\u7d66\u7684\u662f\u96a8 , \u03a9 3.1)\u70ba\u6a19\u6e96\u7684\u5377\u7a4d\ufa19\u7d93\u7db2\uf937\u67b6\u69cb\u4e4b\u7bc4\uf9b5\u3002\u4ee5\u5716\u7247\u5206\uf9d0\u70ba\uf9b5\uff0c \u65e2\u80fd\u4fdd\u6709\u8a9e\u97f3\u4e2d\u7684\u91cd\u8981\u8cc7\u8a0a\u53c8\u80fd\u6709\u6548\u7684\ufa09\u4f4e\u8f38\u5165\u7dad\ufa01(\u5373\u8f38\u5165\u7dad\ufa01\u70ba 2200 \u9ede)\u3002\u5728\u4e00\u7dad\u7684\u5377\u7a4d \u6a5f\u521d\u59cb\u503c\u3002\u6211\u5011\u6700\u5f8c\u6703\u5c0d\u6240\u8a13\uf996\u51fa\u7684\u4e8c\u7dad\u5377\u7a4d\u6838\u9032\ufa08\u5206\u6790\u8207\u8a0e\uf941\u3002 \u800c\u5728\u5716(4.3)\u4e2d\uff0c\u6211\u5011\u5c07\u4e00\u4e9b\u91cd\u8907\u65bc\u591a\u500b\u6a21\u578b\u4e2d\u529f\u80fd\uf9d0\u4f3c\u7684\u5377\u7a4d\u6838\u5708\u51fa\uff0c\u6211\u5011\u53ef\u4ee5\u767c\u73fe\u7121\uf941 III \u7b2c\u4e94\u7a2e\u6a21\u578b Bothrand \u7684\u7d50\u679c\u8a0e\uf941</td></tr><tr><td>| \u6211\u5011\u7684\u8f38\u5165\u53ef\u4ee5\u662f\u4e00\u5f35\u4e8c\u7dad\u7684\u539f\u59cb\u5716\u50cf\uff0c\u5728\u5377\u7a4d\u5c64\u4e2d\u7d93\u904e\u8207\u5377\u7a4d\u6838(kernel)\u7684\u904b\u7b97\u5f8c\uff0c\u53ef\u4ee5\u63d0 ; \u2297 ; \u03a9 |, 0 ; \u03a9 \u03c0 0, \u5c64\u6642\uff0c\u70ba\uf9ba\u80fd\u5b8c\u6574\u7684\u8868\u73fe\u5176\uf984\u6ce2\u5668\u7684\u5927\u5c0f\u80fd\u5920\u6db5\u84cb\u5404\u7a2e\u983b\uf961\uff0c\u56e0\u6b64\u9078\u64c7\u5377\u7a4d\u6838\u5927\u5c0f\u70ba 25ms (200 \u5728\u4f55\u7a2e\u6a21\u578b\u4e2d\uff0c\ufa26\u5b58\u5728\u8457\uf9d0\u4f3c\u529f\u80fd\u7684\u5377\u7a4d\u6838\uff0c\u9019\uf96f\u660e\uf9ba\u7121\uf941\u4e8c\u7dad\u5377\u7a4d\u6838\u662f\u5426\u6709\u7d66\u5b9a\u521d\u59cb\u503c\uff0c\u7d93 \u6839\u64da\u8868(4.2)\uff0c\u6211\u5011\u53ef\u4ee5\u767c\u73fe Bothrand \u6a21\u578b\u7684\u8868\u73fe\uf967\u5982\u5176\u4ed6\u6709\u7d66\u5b9a\u521d\u59cb\u5f62\uf9fa\u7684\u6a21\u578b\uff0c\u6211\u5011 (2.7) , , \u03a9 | ; \u2297 ; \u03a9 |, 0; 0 \u03a9 \u03c0 0, \u53d6\u5230\u5176\u76f8\u5c0d\u61c9\u7684\u7279\u5fb5\u5716(feature map)\uff0c\u6bcf\u500b\u5377\u7a4d\u6838\u6240\u5f97\u5230\u7684\u7279\u5fb5\u5716\u7686\u70ba\u4e00\u7368\uf9f7\u5e73\u9762\uff0c\u4e14\u5176\u5e73\u9762 \u4e0a\u6240\u6709\ufa19\u7d93\u5143\u4e4b\u6b0a\u503c\u76f8\u7b49\uff0c\u6b64\u6b65\u9a5f\u65bc\u7269\uf9e4\u610f\u7fa9\u4e0a\u70ba\u63d0\u53d6\u8207\u76ee\u6a19\u76f8\u95dc\u4e4b\u7279\u5fb5\uff0c\u4ee5\uf9dd\u6211\u5011\u4e4b\u5f8c\u7684\u8a08 \u7b97\u3002\u5716(3.2)\u70ba\u5377\u7a4d\u5c64\uf969\u5b78\u904b\u7b97\u4e4b\u7bc4\uf9b5\u3002 \u9ede)\uff0c\u6b64\u8a2d\u5b9a\u53ef\u4ee5\u6a21\u64ec\u4e2d\u5fc3\u983b\uf961\u70ba 80Hz ~ 4000Hz \u7684\u5e36\u901a\uf984\u6ce2\u5668\u7684\u8108\u885d\u97ff\u61c9\uff0c\u4e26\u4e14\u4ee5\u6bcf 10ms (80 \u9ede)\u70ba\u97f3\u6846\u5f7c\u6b64\u9593\u7684\u9593\u9694\u3002\u85c9\u6b64\uf92d\u6a21\u4eff\u5728\u505a\u983b\u8b5c\u5206\u6790\u6642\uff0c\u539f\u59cb\u8a0a\u865f\u6642\u9593\u8ef8\u4e0a\u7684\u8655\uf9e4\u65b9\u5f0f\u3002 \u7d93\u904e\u4e00\u7dad\u5377\u7a4d\u5c64\u5f8c\uff0c\u6211\u5011\u5c07 20 \u500b\uf984\u6ce2\u5668\u7d50\u679c\u6392\u6210\u4e00\u5f35\u983b\u8b5c\u5716(\u5927\u5c0f\u70ba\uff1a20 kernel * n frame)\uff0c \u5716 3.5\uff1a20 \u500b\u4e00\u7dad\u5377\u7a4d\u6838\u7d93\u904e\u5085\uf9f7\uf96e\u8f49\u63db\u6240\u5f97\u4e4b\u983b\uf961\u632f\u5e45\u97ff\u61c9 Model SNR(dB) \u904e\u5927\u8cc7\uf9be\u7684\u8a13\uf996\u5f8c\ufa26\u6703\u6f14\u5316\u751f\u6210\u51fa\uf9d0\u4f3c\u7684\u5377\u7a4d\u6838\uff0c\u800c\u5c0e\u81f4\u9019\u4e9b\u6a21\u578b\u7684\u6700\u7d42\u7d50\u679c\u5dee\u8ddd\uf967\u5927\u3002 \u731c\u60f3\u53ef\u80fd\u539f\u56e0\u662f Bothrand \u6a21\u578b\u7684\u5377\u7a4d\u6838\u53ef\u80fd\u9084\u9700\u8981\u8f03\u9577\u7684\u6642\u9593\u6216\u8f03\u591a\u7684\u8cc7\uf9be\u624d\u80fd\u8a13\uf996\u51fa\uf901\u6709 Gammainit_A1init\uff1a\u4e00\u7dad\u5377\u7a4d\u6838\u7684\u521d\u59cb\u5f62\uf9fa\u7d66\u5b9a\u70ba 20 \u500b gammatone \uf984\u6ce2\u5668\u4e4b\u7d50\u679c\u3001\u4e8c\u7dad\u5377\u7a4d 1D CNN kernel 2D CNN kernel -5 0 5 \u6548\u679c\u7684\u5f62\uf9fa\uff0c\u5728\u6b64\uff0c\u6211\u5011\u50c5\u5c31\u73fe\u968e\u6bb5\u7684\u7d50\u679c\u9032\ufa08\uf96f\u660e\u3002 \u6838\u7684\u521d\u59cb\u5f62\uf9fa\u7d66\u5b9a\u70ba\u8a08\u7b97\u51fa\uf92d\u4e4b 24 \u500b STMF\uff0c\u5f8c\uf92d\u900f\u904e\u524d\u994b\u4ee5\u53ca\u53cd\u5411\u50b3\u64ad\u6f14\u7b97\u6cd5\u5c0d\uf978\u968e\u6bb5\u7684\u5377 \u7a4d\u6838\u9032\ufa08\u8abf\u6574\u3002\u9019\u500b\u6a21\u578b\u7684\u5047\u8a2d\u662f\uff0c\u807d\u89ba\u611f\u77e5\u6a21\u578b\u7684\uf978\u500b\u968e\u6bb5\u7684\ufa19\u7d93\u53cd\u61c9\u7686\u53ef\u4ee5\u91dd\u5c0d\u61c9\u7528\u76ee\u7684 \u7684\uf967\u540c\u800c\u9032\ufa08\u8abf\u6574\u3002 A1 initial 59.50% 77.25% \u5716 4.4\uff1a\u64f7\u53d6\u8abf\u8b8a\u983b\uf961\u8b8a\u5316\u70ba 4Hz \u7684\u5377\u7a4d\u6838 95.00% Gammatone Fix A1 random 63.50% 73.25% 93.75% \u4e0b\u5716(4.7)\u6211\u5011\u53ef\u4ee5\u770b\u5230 Bothrand \u6a21\u578b\u7684\u4e00\u7dad\u5377\u7a4d\u6838\u983b\uf961\u632f\u5e45\u97ff\u61c9\u5716\uff0c\uf967\u540c\u65bc\u5148\u524d\u7684\u6bd4\u8f03 (2.8) \u800c \u4ee3\u8868\u4e00\u7dad\u7684\u5085\uf9f7\uf96e\u8f49\u63db\uff0c\u2297 \u662f\u5916\u7a4d\u3002rate (\u03c9, \u4ee5 Hz \u70ba\u55ae\u4f4d) \u548c scale (\u03a9, \u4ee5 ms \u70ba\u55ae \u4f4d) \u5206\u5225\u662f\u6642\u9593\u7684\u983b\u57df\u8ef8\u4ee5\u53ca\u983b\uf961\u7684\u983b\u57df\u8ef8\u3002\u800c \u548c \u5247\u662f\u4ee3\u8868\uf9dd\u7528\u73c8\u746a\u5f62\uf9fa\uf984\u6ce2\u5668 (Gammatone filters)\u6240\u5f97\u5230\u7684\u4ee5 \u53ca\u03a9 \u70ba\u4e2d\u5fc3\u7684\u4e00\u7dad\u6642\u9593\u53ca\u983b\uf961\u8108\u885d\u97ff\u61c9\uff0c\u5982(2.9)\u3002 ; cos 2 ; \u03a9 cos 2 \u03a9 (2.9) \u5716 3.2\uff1a\u5377\u7a4d\u6838\u5927\u5c0f\u70ba 3x3 \u4e4b\u5377\u7a4d\u5c64\u7bc4\uf9b5 \u5716 3.4\uff1a\u6240\u63d0\u51fa\u7684\u6a21\u578b\u67b6\u69cb \u85c9\u7531\u807d\u89ba\u611f\u77e5\u6a21\u578b\u7684\u555f\u767c\uff0c\u6211\u5011\u63d0\u51fa\uf9ba\u4e00\u500b\u57fa\u65bc\u5377\u7a4d\ufa19\u7d93\u7db2\uf937\u7684\u8a9e\u8005\uf9fc\u5225\u7cfb\u7d71\u3002\u6211\u5011\u6240\u63d0 \u51fa\u7684\uf9d0\ufa19\u7d93\u6a21\u578b\uff0c\u5305\u542b\uf9ba\u8f38\u5165\u5c64\u3001\u4e00\u7dad\u7684\u5377\u7a4d\u5c64\u3001\u4e8c\u7dad\u7684\u5377\u7a4d\u5c64\u3001\u6c60\u5316\u5c64\uff0c\u4ee5\u53ca\u56db\u5c64\u7279\u5fb5\u6574\u5408 \u5c64\uff0c\u5982\u5716(3.4)\u6240\u793a\u3002\u5176\u4e2d\uff0c\u70ba\uf9ba\u8981\u5b8c\u6574\u7684\u6a21\u64ec\u807d\u89ba\u611f\u77e5\u6a21\u578b\uff0c\u6211\u5011\u8f38\u5165\u5c64\u7684\u662f\u672a\u7d93\u904e\u4efb\u4f55\u8655\uf9e4 \u7684\u4e00\u7dad\u539f\u59cb\u97f3\u6a94\u3002 \u5728\u4e00\u7dad\u7684\u5377\u7a4d\u5c64\u6642\uff0c\u6211\u5011\uf9dd\u7528\uf9ba\u5377\u7a4d\u6838\u6b0a\u503c\u5171\u4eab\u7684\u7279\u6027\uff0c\u6211\u5011\u8a8d\u70ba\u5728\u5c0d\u539f\u59cb\u97f3\u6a94\u505a\u5377\u7a4d\u6642\uff0c \u76f8\u7576\u65bc\u5c0d\u5176\u505a\uf9ba\uf967\u540c\u983b\uf961\u7684\uf984\u6ce2\u3002\u56e0\u6b64\u6211\u5011\u6839\u64da\u8033\u8778\u5c0d\u65bc\uf967\u540c\u4e2d\u5fc3\u983b\uf961\u4ee5\u53ca\u983b\u5bec\u7684\u5e38\uf969 Q \u95dc \u8003\u616e\u5230\u4e8c\u7dad\u5377\u7a4d\u6838\u7684\u7269\uf9e4\u610f\u7fa9\uff0c\u5728\u6642\u9593\u8ef8\u4e0a\uff0c\u5224\u65b7\u4e00\u500b\u97f3\u7d20\u6700\u5c11\u9808 50ms \u7684\u6642\u9593\uff0c\u6211\u5011\u8a2d\u8a08\u7684 3.6 \u4e8c\u7dad\u5377\u7a4d\u6838\u521d\u59cb\u5316 Gammainit_A1rand\uff1a\u4e00\u7dad\u5377\u7a4d\u6838\u7684\u521d\u59cb\u5f62\uf9fa\u7d66\u5b9a\u70ba 20 \u500b gammatone \uf984\u6ce2\u5668\u4e4b\u7d50\u679c\uff1b\u800c\u4e8c\u7dad\u5377 \u7a4d\u6838\uf967\u7d66\u7279\u5b9a\u7684\u521d\u59cb\u503c\uff0c\u76f4\u63a5\u900f\u904e\u524d\u994b\u4ee5\u53ca\u53cd\u5411\u50b3\u64ad\u6f14\u7b97\u6cd5\u9032\ufa08\u8a13\uf996\u3002\u9019\u500b\u6a21\u578b\u7684\u5047\u8a2d\u8207\u7b2c\u4e09 \uf9d0\u578b(Gammainit_A1init)\u76f8\u4f3c\uff0c\uf967\u540c\u7684\u5730\u65b9\u662f\u4e8c\u7dad\u5377\u7a4d\u6838\u7d66\u7684\u662f\u96a8\u6a5f\u521d\u59cb\u503c\u3002\u6211\u5011\u6700\u5f8c\u6703\u5c0d\u6240 \u8a13\uf996\u51fa\u7684\u4e8c\u7dad\u5377\u7a4d\u6838\u9032\ufa08\u5206\u6790\u8207\u8a0e\uf941\u3002 A1 initial 67.00% 77.50% \u540c\u6a23\u7684\u6211\u5011\u4e5f\u53ef\u4ee5\u5f9e\u5377\u7a4d\u6838\u627e\u51fa\u5176\u4ed6\u8abf\u8b8a\u983b\uf961\u8b8a\u5316\uff0c\u5982 8\u300116\u300132Hz \u7684\u6ce2\u578b\uff0c\u5982\u4e0b\u5716(4.5) \u6a21\u578b\uff0cBothrand \u56e0\u70ba\u662f\u96a8\u6a5f\u7d66\u5b9a\u521d\u59cb\u503c\u800c\u76f4\u63a5\u9032\ufa08\u8a13\uf996\uff0c\u56e0\u6b64\u5176\u4e26\u6c92\u6709\u50cf\u6211\u5011\u5148\u524d\u7528\uf92d\u7d66\u5b9a\u521d 92.00% Gammatone Initial A1 random 69.25% 76.50% 92.75% Both random 56.00% 65.75% 87.00% \u8868 4.2\uff1a\u5404\u6a21\u578b\u5728\u591a\u8a0a\u96dc\u6bd4\u8207\u591a\u96dc\u8a0a\u7a2e\uf9d0\u689d\u4ef6\u4e0b\u7684\u8a9e\u8005\uf9fc\u5225\u6b63\u78ba\uf961 \u5716 4.1\uff1a\u591a\u8a0a\u96dc\u6bd4\u53ca\u591a\u96dc\u8a0a\u7a2e\uf9d0\u689d\u4ef6\u4e0b\uff0c\u6a21\u578b\u8a13\uf996\u5f8c\u4e4b\u4e00\u7dad\u5377\u7a4d\u6838\u983b\uf961\u632f\u5e45\u97ff\u61c9\u5716 \u6240\u793a\u3002\u7576\u7136\uff0c\u6240\u6709\u5377\u7a4d\u6838\u4ee3\u8868\u7684\u8abf\u8b8a\u983b\uf961\u8b8a\u5316\uf967\u55ae\u55ae\u53ea\u6709\u9019\u4e9b\uff0c\u56e0\u6b64\u6211\u5011\u63a8\u65b7\uff0c\u5728\u8a9e\u8005\u8fa8\uf9fc\u9019 \u500b\u8b70\u984c\u4e0a\u8abf\u8b8a\u983b\uf961\u8b8a\u5316\u662f\u4e00\u9805\u91cd\u8981\u7684\u8cc7\u8a0a\u3002 \u59cb\u503c\u7684 gammatone \uf984\u6ce2\u5668\u6709\u4f9d\u7167\u4e2d\u5fc3\u983b\uf961\uf92d\u6392\u5b9a\u5927\u5c0f\u9806\u5e8f\uff0c\u56e0\u800c\u5448\u73fe\u51fa\u4e00\u7d44\u6c92\u6709\u898f\u5247\u7684\uf984\u6ce2 \u5668\u7d44\u3002\u4f46\u6211\u5011\u53ef\u4ee5\u5f9e\u8a72\u983b\uf961\u97ff\u61c9\u5716\u4e2d\u767c\u73fe\uff0c\u5176\u5c0d\u65bc\uf967\u540c\u983b\uf961\u4ecd\u6703\u6709\uf967\u540c\u7684\u89e3\u6790\u6548\u679c\uff0c\u5c31\u5982\u540c gammatone \uf984\u6ce2\u5668\u5728\u4f4e\u983b\u6642\u89e3\u6790\u8f03\u70ba\ufa1d\u7d30\uff0c\u800c\u9ad8\u983b\u6642\u89e3\u6790\u5247\u8f03\u5dee\u3002 \u5716 4.8\uff1a Bothrand \u6a21\u578b\u7684\u4e8c\u7dad\u5377\u7a4d\u6838\u5f62\uf9fa\u7d50\u679c\u5716 \u5377\u7a4d\u6838 y 3.5 \u4e00\u7dad\u5377\u7a4d\u6838\u521d\u59cb\u5316 Bothrand\uff1a\u4e00\u7dad\u8207\u4e8c\u7dad\u5377\u7a4d\u6838\uff0c\u7686\uf967\u7d66\u5b9a\u7279\u5b9a\u521d\u59cb\u503c\uff0c\u76f4\u63a5\u900f\u904e\u524d\u994b\u4ee5\u53ca\u53cd\u5411\u50b3\u64ad\u6f14\u7b97\u6cd5\u9032\ufa08 \u8a13\uf996\u3002\u6211\u5011\u60f3\u85c9\u7531\uf967\u7d66\u5b9a\u4efb\u4f55\u521d\u59cb\u503c\u7684\uf9fa\u6cc1\uff0c\uf92d\u89c0\u5bdf\u5728\u6b64\u67b6\u69cb\u4e0b\u8a13\uf996\u8abf\u6574\u5b8c\u5f8c\u4e00\u7dad\u53ca\u4e8c\u7dad\u5377\u7a4d I 5. \u7d50\uf941\u8207\u672a\uf92d\u5c55\u671b II \u524d\u56db\u7a2e\u6a21\u578b\uff0c\u5c0d\u4e8c\u7dad\u5377\u7a4d\u6838\u5f62\uf9fa\u9032\ufa08\u8a0e\uf941 \u524d\u56db\u7a2e\u6a21\u578b\uff0c\u5c0d\u4e00\u7dad\u5377\u7a4d\u6838\u7684\u5f62\uf9fa\u9032\ufa08\u8a0e\uf941 \u9996\u5148\uff0c\u6211\u5011\u5148\u91dd\u5c0d\u524d\u56db\u7a2e\u6a21\u578b\uff0c\u4e5f\u5c31\u662f\u6709\uf96b\u8003 gammatone \uf984\u6ce2\u5668\u7684\u6a21\u578b\uff0c\u5206\u6210\u56fa\u5b9a\u5176\u503c \u7531\u4e0a\u8ff0\u8a0e\uf941\u6211\u5011\u53ef\u4ee5\u77e5\u9053\uff0c\u524d\u56db\u7a2e\u6a21\u578b\u7121\uf941\u662f\u5426\u900f\u904e\u8a13\uf996\u9032\ufa08\u4fee\u6b63\uff0c\u4e00\u7dad\u5377\u7a4d\u6838\ufa26\u5927\u81f4\u4e0a \u5728\u672c\uf941\u6587\uf9e8\uff0c\u6211\u5011\u63d0\u51fa\uf9ba\u4e00\u500b\u57fa\u65bc\uf978\u968e\u6bb5\u4e4b\u807d\u89ba\u611f\u77e5\u6a21\u578b\u4e4b\uf9d0\ufa19\u7d93\u7db2\uf937\u7684\u6a21\u578b\uff0c\u4e26\u5c07\u5176\u61c9</td></tr></table>" | |
| } | |
| } | |
| } | |
| } |