| { |
| "paper_id": "O07-1004", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T08:08:33.102896Z" |
| }, |
| "title": "\u52a0\u6210\u6027\u96dc\u8a0a\u74b0\u5883\u4e0b\u904b\u7528\u7279\u5fb5\uf96b\uf969\u7d71\u8a08\u88dc\u511f\u6cd5\u65bc\u5f37\u5065\u6027\u8a9e\u97f3\u8fa8\uf9fc Feature Statistics Compensation for Robust Speech Recognition in Additive Noise Environments", |
| "authors": [ |
| { |
| "first": "Tsung-Hsueh", |
| "middle": [], |
| "last": "\u8b1d\u5b97\u5b78", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| }, |
| { |
| "first": "\u570b\uf9f7\u66a8\u5357\u570b\u969b\u5927\u5b78\u96fb\u6a5f\u5de5\u7a0b\u5b78\u7cfb", |
| "middle": [], |
| "last": "Hsieh", |
| "suffix": "", |
| "affiliation": {}, |
| "email": "" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "In this paper, we propose several compensation approaches to alleviate the effect of additive noise on speech features for speech recognition. These approaches are simple yet efficient noise reduction techniques that use online constructed pseudo stereo codebooks to evaluate the statistics in both clean and noisy environments. The process yields transforms for noise-corrupted speech features to make them closer to their clean counterparts. We apply these compensation approaches on various well-known speech features, including mel-frequency cepstral coefficients (MFCC), autocorrelation mel-frequency cepstral coefficients (AMFCC), linear prediction cepstral coefficients (LPCC) and perceptual linear prediction cepstral coefficients (PLPCC). Experimental results conducted on the Aurora-2 database show that the proposed approaches provide all types of the features with a significant performance gain when compared to the baseline results and those obtained by using the conventional utterance-based cepstral mean and variance normalization (CMVN).", |
| "pdf_parse": { |
| "paper_id": "O07-1004", |
| "_pdf_hash": "", |
| "abstract": [ |
| { |
| "text": "In this paper, we propose several compensation approaches to alleviate the effect of additive noise on speech features for speech recognition. These approaches are simple yet efficient noise reduction techniques that use online constructed pseudo stereo codebooks to evaluate the statistics in both clean and noisy environments. The process yields transforms for noise-corrupted speech features to make them closer to their clean counterparts. We apply these compensation approaches on various well-known speech features, including mel-frequency cepstral coefficients (MFCC), autocorrelation mel-frequency cepstral coefficients (AMFCC), linear prediction cepstral coefficients (LPCC) and perceptual linear prediction cepstral coefficients (PLPCC). Experimental results conducted on the Aurora-2 database show that the proposed approaches provide all types of the features with a significant performance gain when compared to the baseline results and those obtained by using the conventional utterance-based cepstral mean and variance normalization (CMVN).", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Abstract", |
| "sec_num": null |
| } |
| ], |
| "body_text": [ |
| { |
| "text": "\u23aa \u23aa \u23aa \u23aa \u23aa \u23aa \u23aa \u23aa = \u0394 \u0394 \u23a8 \u23ac \u23aa \u23aa \u23aa \u23aa \u23aa \u23aa \u0394 \u0394 \u23aa \u23aa \u23aa \u23aa \u23a9 \u23ad", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u9810\u5f37\u8abf \u6f22\u660e\u8996\u7a97 \u5012\u983b\u8b5c\u8f49\u63db { } { } 2 i i i i c c c c \u23a7 \u23ab \u23aa \u23aa \u23aa \u23aa = \u0394 \u23a8 \u23ac \u23aa \u23aa \u0394 \u23aa \u23aa \u23a9 \u23ad i c", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u4e9b\u6539\u8b8a\u90fd\u662f\u91dd\u5c0d\u4eba\u7684\u807d\u89ba\u7279\u6027\u800c\u505a\u7684\u3002 \u9810\u5f37\u8abf \u6f22\u660e\u8996\u7a97 \uf9ea\u6563\u9918\u5f26\u8f49\u63db { } { } 2 i i i i \u23a7 \u23ab \u23aa \u23aa \u23aa \u23aa \u23aa \u23aa \u23aa \u23aa \u23aa \u23aa = \u0394 \u23a8 \u23ac \u23aa \u23aa \u23aa \u23aa \u23aa \u23aa \u0394 \u23aa \u23aa \u23aa \u23aa \u23a9 \u23ad", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u4e7e\u6de8\u7684\u8a9e\u97f3\u8a0a\u865f\u5728\u7d93\u904e\u52a0\u6210\u6027\u96dc\u8a0a\u5e72\u64fe\u5f8c\uff0c\u5176\u5012\u983b\u8b5c\u4e4b\u5e73\u5747\u503c\u6703\u548c\u539f\u672c\u7684\u4e7e\u6de8\u8a9e\u97f3\u5012\u983b\u8b5c \u7684\u5e73\u5747\u503c\u4e4b\u9593\u6703\u5b58\u5728\u4e00\u500b\u504f\u79fb\uf97e\uff0c\u800c\u5176\u8b8a\uf962\uf969\u76f8\u5c0d\u65bc\u4e7e\u6de8\u8a9e\u97f3\u7279\u5fb5\uf96b\uf969\u800c\u8a00\u5247\u6703\u6709\u58d3\u7e2e\u6027\uff0c\u56e0 \u6b64\u6703\u9020\u6210\u8a13\uf996\u8207\u6e2c\u8a66\u7279\u5fb5\u7684\uf967\u5339\u914d\u800c\ufa09\u4f4e\u8fa8\uf9fc\u6548\u679c\u3002\u800c\u4f7f\u7528\u5012\u983b\u8b5c\u5e73\u5747\u503c\u8207\u8b8a\uf962\uf969\u7b49\u5316\u6cd5 [1](CMVN)\u53ef\u5c07\u6bcf\u4e00\u7dad\u5012\u983b\u8b5c\u7279\u5fb5\uf96b\uf969\u4e4b\u5e73\u5747\u503c\u5316\u70ba\uf9b2\uff0c\u4e26\u5c07\u5176\u8b8a\uf962\uf969\u6b63\u898f\u5316\u70ba 1\uff0c\u9019\u6a23\u5c31\u80fd \ufa09\u4f4e\u4e0a\u8ff0\u6240\u8b02\u7684\u504f\u79fb\uf97e\u8207\u58d3\u7e2e\u6027\uff0c\u9032\u800c\u63d0\u5347\u5012\u983b\u8b5c\uf96b\uf969\u7684\u5f37\u5065\u6027\u3002 \u6574\u6bb5\u5f0f\u5012\u983b\u8b5c\u5e73\u5747\u503c\u8207\u8b8a\uf962\uf969\u7b49\u5316\u6cd5\u7684\u4f5c\u6cd5\u5982(\u5f0f 3-1)\uff0c\u5047\u8a2d [ ] { } , 1,2, , Y n n N = \u2026 \u70ba\u4e00\u7531\u8a9e \u97f3\u8cc7\uf9be\u64f7\u53d6\u6240\u5f97\u5230\u7684\u67d0\u4e00\u7dad\u5012\u983b\u8b5c\u7279\u5fb5\uf96b\uf969\u5e8f\uf99c\uff0c\u800c\u7d93\u904e\u6574\u6bb5\u5f0f\u5012\u983b\u8b5c\u5e73\u5747\u503c\u8207\u8b8a\uf962\uf969\u7b49\u5316\u6cd5 \u8655\uf9e4\u5f8c\uff0c\u5f97\u5230\u65b0\u7684\u7279\u5fb5\uf96b\uf969 [ ] { } , 1,2, , U CMVN Y n n N \u2212 = \u2026 \uff0c\u5176\u4e2d\u7684 [ ] { } , 1,2, , U CMVN Y n n N \u2212 = \u2026 \u5e73\u5747\u503c \u8207\u6a19\u6e96\u5dee\u662f\u7d93\u7531\u6574\u6bb5\u8a9e\u97f3\u7684\u97f3\u6846\u6c42\u53d6\u800c\u5f97\uff0c\u5982\u5f0f(3-2)\u8207\u5f0f(3-3)\u3002 [ ] [ ] , 1, 2,...., Y U CMVN Y Y n Y n n N \u03bc \u03c3 \u2212 \u2212 = = ( \u5f0f 3 -1 ) \u5176\u4e2d 1 1 [ ] N Y n Y n N \u03bc = = \u2211 ( \u5f0f 3 -2 ) 2 1 1 ( [ ] ) N Y Y n Y n N \u03c3 \u03bc = = \u2212 \u2211 ( \u5f0f 3 -3 ) ( \u4e8c ) \u5206 \u6bb5 \u5f0f \u5012 \u983b \u8b5c \u5e73 \u5747 \u503c \u8207 \u8b8a \uf962 \uf969 \u6b63 \u898f \u5316 \u6cd5 (", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u5716 \uf9d1 (a) \u662f \u5f9e AURORA2 \u4e2d \u4e7e \u6de8 \u7684 \u8a13 \uf996 \u8a9e \uf9be \u5eab \u4e2d \uff0c \u4e09 \u6bb5 \uf967 \u540c \u8a9e \u97f3 \"FAC_1911446\" \uff0c \"FAC_1473533A\"\u8207\"FAC_1O1\"\u64f7\u53d6\u51fa\u8072\u5b78\u55ae\u4f4d\"one\"\u4e4b\u539f\u59cb\u7b2c\u4e00\u7dad\u5012\u983b\u8b5c\uf96b\uf969 c1 \u4e4b\uf9d7\ufa0b\uff1b\u5716\uf9d1 (b)\u5247\u662f\u5716\uf9d1(a)\u7d93\u904e U-CMVN \u8655\uf9e4\u5f8c\u7684\u7248\u672c\u3002\u5f9e\u5716\uf9d1(a)\u53ef\u767c\u73fe\u672a\u7d93 CMVN \u8655\uf9e4\u7684 c1\uff0c\u5176\uf9d7\ufa0b \u5728([-10.3\uff0c12.8]\uff0c[-11.6\uff0c13.0]\u53ca[-11.0\uff0c15.0])\u9019\u4e09\u500b\u7bc4\u570d\u5167\u6709\u76f8\u4f3c\u7684\u5206\u4f48\uf9fa\u6cc1\uff1b\u53cd\u89c0\u5716\uf9d1(b)\uff0c \u56e0\u70ba\uf967\u540c\u8a9e\u97f3\u4e2d\u7684\u7279\u5fb5\uf96b\uf969\u662f\u88ab\uf967\u540c\u7684\u5e73\u5747\u503c\u8207\u8b8a\uf962\uf969\u4f5c\u6b63\u898f\u5316\uff0c\u4f7f\u6b63\u898f\u5316\u5f8c\u4e09\u500b c1 \u7684\uf9d7\ufa0b\u8b8a \u5f97\uf967\u592a\u76f8\u540c\uff0c\u5b83\u5011\u7684\u5206\u4f48\uf9fa\u6cc1\u5728([-0.5\uff0c2.0]\uff0c[-0.5\uff0c2.8]\u53ca[-1.1\uff0c2.3])\uff0c\u9019\u4e09\u500b\u7bc4\u570d\u8ddf\u4e4b\u524d\u76f8\u6bd4 \u4e0b\u662f\u6bd4\u8f03\uf967\u540c\u7684\u3002 (a) (b)", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u5716\uf9d1\uff1aAURORA2 \u4e2d\u4e7e\u6de8\u7684\u8a13\uf996\u8a9e\uf9be\u5eab\u4e2d\uff0c\u4e09\u6bb5\uf967\u540c\u8a9e\u97f3\"FAC_1911446\"\uff0c\"FAC_1473533A\" \u8207\"FAC_1O1\"\u64f7\u53d6\u51fa\u8072\u5b78\u55ae\u4f4d\"one\"\u4e4b(a)\u539f\u59cb\u7b2c\u4e00\u7dad\u6885\u723e\u5012\u983b\u8b5c\u7279\u5fb5 c1 ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\uf9d7\ufa0b(b)\u7d93 U-CMVN \u8655 \uf9e4\u5f8c\u7b2c\u4e00\u7dad\u6885\u723e\u5012\u983b\u8b5c\u7279\u5fb5 c1", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "%[ ] f x n % ( [ ]) f y m = + % % % [ ] [ ] [ ] y m x n n p [ ] y m i y ( ) cep i y x \u5716\u4e03\uff1a\u865b\u64ec\u96d9\u901a\u9053\u78bc\u7c3f\u4e4b\u5efa\uf9f7\u67b6\u69cb\u53ca\u4ee5\u865b\u64ec\u96d9\u901a\u9053\u78bc\u7c3f\u57f7\ufa08\u7279\u5fb5\uf96b\uf969\u88dc\u511f\u6cd5\u4e4b\uf9ca\u7a0b\u5716 \u865b\u64ec\u96d9\u901a\u9053\u78bc\u7c3f\u4e4b\u5efa\uf9f7\u904e\u7a0b\u8a73\u8ff0\u5982\u4e0b\uff1a \u9996\u5148\u5c07\u8a9e\uf9be\u5eab\u4e2d\u6240\u6709\u4e7e\u6de8\u8a9e\uf9be\u7684\u6bcf\u4e00\u6bb5\u8a9e\u97f3\uff0c\u900f\u904e\u7279\u5fb5\uf96b\uf969\u64f7\u53d6\uf9ca\u7a0b\u8f49\u63db\u6210\u4e00\u5e8f\uf99c\u7684\u4e2d\u4ecb \u7279\u5fb5\u5411\uf97e\uff0c\u5982\u8868\u4e00\u6240\u8ff0\u3002\u9019\u4e9b\u7531\u6240\u6709\u4e7e\u6de8\u8a9e\uf9be\u7684\u8a9e\u8abf\u6240\u5f97\u5230\u7684\u4e2d\u4ecb\u7279\u5fb5\u5411\uf97e\uff0c\u900f\u904e\u5411\uf97e\uf97e\u5316 (vector quantization\uff0cVQ)\u5f8c\u53ef\u5efa\uf9f7\u6210\u4e00\u7d44\u5305\u542b N \u500b\u78bc\u5b57(codewords)\u7684\u96c6\u5408\uff0c\u4ee5 { [ ],1 } n n N \u2264 \u2264 x \u8868 \u793a\u3002\u9019\u7d44\u5728\u4e2d\u4ecb\u7279\u5fb5\uf96b\uf969\u57df\u4e0a\u7684\u4e7e\u6de8\u8a9e\u97f3\u78bc\u7c3f\u4e2d\u6240\u6709\u78bc\u5b57\uff0c\u90fd\u53ef\u7d93\u904e\u5269\u4e0b\u7684\u7279\u5fb5\uf96b\uf969\u64f7\u53d6\u6b65\u9a5f \u8f49\u63db\u81f3\u5012\u983b\u8b5c\u57df\uff0c\u5982(\u5f0f 4-1)\u6240\u793a\uff1a ( ) [ ] [ ] n f n = x x \uff0c ( \u5f0f 4 -1 ) \u8868\u4e00\uff1a\u672c\uf941\u6587\u4e2d\u6240\u4f7f\u7528\u4e4b\u56db\u7a2e\u8a9e\u97f3\u7279\u5fb5\uf96b\uf969\uff0c\u53ca\u5176\u5177\u5099\u8a9e\u97f3\u8207\u96dc\u8a0a\u7dda\u6027\u76f8\u52a0\u7279\u6027\u4e4b\u4e2d\u4ecb\u7279\u5fb5\uf96b", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "\u5176\u4e2d ( ) . f \u70ba\u8f49\u63db\u51fd\uf969\uff0c\u5b83\u662f\u96a8\u8457\u6211\u5011\u6240\u9078\u64c7\u7684\u7279\u5fb5\uf96b\uf969\u578b\u614b\u800c\uf967\u540c\u3002\u56e0\u6b64 { } [ ],1 n n N \u2264 \u2264 x \u9019\u7d44 \u7d93\u8f49\u63db\u81f3\u5012\u983b\u8b5c\u57df\u7684\u78bc\u7c3f\uff0c\u5373\u7a31\u4e4b\u70ba\u4e7e\u6de8\u8a9e\u97f3\u7684\u5012\u983b\u8b5c\u78bc\u7c3f\u3002 \u81f3\u65bc\u542b\u96dc\u8a0a\u7684\u6e2c\u8a66\u8a9e\u97f3\u65b9\u9762\uff0c\u56e0\u70ba\u8981\u5b8c\u5168\u4ee5\u6bcf\u6bb5\u7c21\u77ed\u7684\u6e2c\u8a66\u8a9e\u97f3\u70ba\u57fa\u790e\uff0c\u53bb\u5efa\uf9f7\u4e00\u7d44\u53ef\u9760 \u7684\u78bc\u5b57\u662f\u5f88\u56f0\u96e3\u7684\uff0c\u6240\u4ee5\u6211\u5011\u8a66\u8457\u85c9\u7531\u4e7e\u6de8\u8a9e\u97f3\u5728\u4e2d\u4ecb\u7279\u5fb5\uf96b\uf969\u57df\u4e0a\u7684\u78bc\u5b57\uff0c\uf92d\u5efa\uf9f7\u5c0d\u61c9\u81f3\u8a72 \u6bb5\u7684\u542b\u96dc\u8a0a\u4e4b\u6e2c\u8a66\u8a9e\u97f3\u7684\u78bc\u7c3f\u3002\u6b65\u9a5f\u5982\u4e0b\uff1a \u5c0d\u65bc\u4e00\u6bb5\u6e2c\u8a66\u8a9e\u97f3\uff0c\u6211\u5011\u5047\u8a2d\u4f30\u8a08\u5230\u7684\u7d14\u96dc\u8a0a\u5728\u4e2d\u4ecb\u7279\u5fb5\uf96b\uf969\u57df\u4e0a\u53ef\u7528\u4e00\u7d44\u5411\uf97e\uf92d\u4ee3\u8868\uff0c \u4ee5 { } [ ],1 p p P \u2264 \u2264 n \u8868\u793a\u3002\u56e0\u70ba\u4e7e\u6de8\u8a9e\u97f3\u8207\u96dc\u8a0a\u5728\u4e2d\u4ecb\u7279\u5fb5\uf96b\uf969\u57df\u4e0a\u662f\u8fd1\u4f3c\u7dda\u6027\u76f8\u52a0\u7684\uff0c\u56e0\u6b64\u542b \u96dc\u8a0a\u8a9e\u97f3\u7684\u78bc\u5b57\u53ef\u8868\u793a\u6210(\u5f0f 4-2)\uff1a ( 1) [ ]| [ ] [ ]", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "EQUATION", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [ |
| { |
| "start": 0, |
| "end": 8, |
| "text": "EQUATION", |
| "ref_id": "EQREF", |
| "raw_str": "\u6211\u5011\uf9dd\u7528\u865b\u64ec\u96d9\u901a\u9053\u78bc\u7c3f\uff0c\u53ef\u4ee5\u7b97\u51fa\u5206\u5225\u4ee3\u8868\u4e7e\u6de8\u8a9e\u97f3\u8207\u96dc\u8a0a\u8a9e\u97f3\u7684\u7d71\u8a08\u503c\uff0c\u5982(\u5f0f 4-4)\u3001 (\u5f0f 4-5)\u6240\u793a\uff1a 2 2 , ,", |
| "eq_num": ", 1 1 1" |
| } |
| ], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "1 ( [ ]) , [( [ ]) ] N N x i i x i i x i n n n n N N \u03bc \u03c3 \u03bc = = \u2248 \u2248 \u2212 \u2211 \u2211 x x (\u5f0f 4-4) 2 2 , , , 1 1 1 1 ( [ ]) , [( [ ]) ] NP NP y i i y i i y i m m m m NP NP \u03bc \u03c3 \u03bc = = \u2248 \u2248 \u2212 \u2211 \u2211 y y (\u5f0f 4-5) \u5176\u4e2d( ) i v \u4ee3\u8868\u4e00\u500b\u4efb\u610f\u5411\uf97e v \u7b2c i \u7dad\u6210\u4efd\uff0c , x i \u03bc \u8207 2 , x i \u03c3 \u5206\u5225\u4ee3\u8868\u4e7e\u6de8\u8a9e\u97f3\u7279\u5fb5\u5411\uf97e x \u7b2c i \u7dad\u7684\u5e73\u5747 \u503c\u8207\u8b8a\uf962\uf969\uff1b , y i \u03bc \u8207 2 , y i \u03c3 \u5206\u5225\u4ee3\u8868\u96dc\u8a0a\u8a9e\u97f3\u7279\u5fb5\u5411\uf97e y \u7b2c i \u7dad\u7684\u5e73\u5747\u503c\u8207\u8b8a\uf962\uf969\u3002\u4ee5\u9019\u4e9b\u7d71\u8a08\u503c\uf92d \u57f7\ufa08\u5012\u983b\u8b5c\u7d71\u8a08\u503c\u88dc\u511f\u6cd5\uff0c\u6211\u5011\u8f49\u63db\u6bcf\u4e00\u6bb5\u96dc\u8a0a\u8a9e\u97f3\u4e4b\u5012\u983b\u8b5c\u5411\uf97e\uff0c\u5982(\u5f0f 4-6)\uff1a ( ) , , , , ( ) [ ] x i i y i x i i y i \u03c3 \u03bc \u03bc \u03c3 = \u00d7 \u2212 + z y (\u5f0f 4-6) \u5728\uf9e4\u60f3\u7684\u60c5\u6cc1\u4e0b\uff0c( ) i z \u8207\u4e7e\u6de8\u8a9e\u97f3\u7279\u5fb5\u5411\uf97e( ) i x \u6703\u6709\u76f8\u540c\u7684\u5e73\u5747\u503c\u8207\u8b8a\uf962\uf97e\uff0c\u7531\u65bc\u96dc\u8a0a\u8a9e\u97f3\u5012\u983b \u8b5c\u7684\u67d0\u4e9b\u7d71\u8a08\u503c\u88ab\u88dc\u511f\uff0c\u4f7f\u5f97\u88dc\u511f\u904e\u5f8c\u7684\u96dc\u8a0a\u8a9e\u97f3\u5012\u983b\u8b5c\u5176\u7d71\u8a08\u503c\u662f\u8fd1\u4f3c\u65bc\u4e7e\u6de8\u8a9e\u97f3\u5012\u983b\u8b5c\u7684 \u7d71\u8a08\u503c\uff0c\u56e0\u6b64\u6211\u5011\u5c07\u6b64\u65b9\u6cd5\u7a31\u70ba\u5012\u983b\u8b5c\u7d71\u8a08\u88dc\u511f\u6cd5(cepstral statistics compensation\uff0cCSC)\u3002\u6211\u5011 \u53ef\u4ee5\u7528\u77e9\u9663\u7684\u5f62\u5f0f\u6539\u5beb(\u5f0f 4-6)\u7684\u5012\u983b\u8b5c\u7d71\u8a08\u88dc\u511f\u6f14\u7b97\u6cd5\uff0c\u5982(\u5f0f 4-7)\uff1a ( ) = \u2212 + y x z y \u03a8 \u03bc \u03bc ( \u5f0f 4 -7 ) \u5176\u4e2d ,1 ,2 ,1 ,2 [ , , ] , [ , ] T T \u03bc \u03bc \u03bc \u03bc = = x x x y y y \u03bc \u03bc \uff0c\u03a8 \u662f\u4e00\u5c0d\u89d2\u7dda\u70ba { } , , / i i \u03c3 \u03c3 x", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "m y \u5c0d\u61c9\u7684\u4e7e\u6de8\u8a9e\u97f3\u78bc\u5b57\u70ba [ ] n x \uff0c\u5176\u4e2d / n m P \u23a1 \u23a4 = \u23a2 \u23a5 ( \u23a1 \u23a4 . \u8868\u793a\u7121\u689d\u4ef6\u9032\u4f4d\u904b\u7b97\uff0c P \u70ba\u7d14\u96dc\u8a0a\u7684\u5411\uf97e\uf969\u76ee)\uff0c { } [ ] n x \u8207 { } [ ] m y \u9019\uf978\u7d44\u78bc\u5b57\u5206\u5225 \u4ee3\u8868\u4e7e\u6de8\u8a9e\u97f3\u8207\u96dc\u8a0a\u8a9e\u97f3\u5012\u983b\u8b5c x \u8207 y \u3002\uf974\u6211\u5011\u80fd\u5c0d\u6bcf\u4e00\u500b\u96dc\u8a0a\u8a9e\u97f3\u78bc\u5b57 [ ] m y \u627e\u5230\u4e00\u500b\u8f49\u63db\u51fd \uf969 ( ) \u22c5 T \uff0c\u4f7f\u5f97 ( ) [ ] m y T \u8207 [ ] n x \u4e4b\u9593\u7684\u6574\u9ad4\u8ddd\uf9ea\u662f\u6700\u5c0f\u7684\uff0c\u90a3\u6211\u5011\u53ef\u4ee5\u5408\uf9e4\u7684\u9810\u671f\u96dc\u8a0a\u8a9e\u97f3\u5012\u983b\u8b5c y \u7d93\u8f49\u63db\u5f8c ( ) y T \uff0c\u6703\uf901\u63a5\u8fd1\u4e7e\u6de8\u8a9e\u97f3\u5012\u983b\u8b5c x\u3002\u70ba\uf9ba\u7c21\u55ae\u8d77\ufa0a\uff0c\u6211\u5011\u5047\u8a2d\u8f49\u79fb\u51fd\uf969\u662f\u57f7\ufa08\u5728 y \u7684 \u6bcf\u4e00\u7dad\u4e0a\u3002\u5047\u8a2d ( ) i \u2022 T \u662f y \u7684\u7b2c i \u7dad\u6210\u4efd\u7684\u8f49\u79fb\u51fd\uf969\uff0c\u5247\u5b9a\u7fa9\u4e00\u76ee\u6a19\u51fd\uf969 i J \u5c07\u4f7f\u5f97 ( ) ( ) [ ] i i m y T \u8207 ( ) [ ] i n x \u7684\u6574\u9ad4\u5e73\u65b9\u8ddd\uf9ea\u6700\u5c0f\uff0c\u5982(\u5f0f 4-8)\uff1a ( ) ( ) ( ) 2 1 [ ] [ ] NP i i i i m J m n = \u23a1 \u23a4 = \u2212 \u23a2 \u23a5 \u23a3 \u23a6 \u2211 y x T (\u5f0f 4-8) \u5176\u4e2d / n m P \u23a1 \u23a4 = \u23a2 \u23a5 \uff0c\u5047\u8a2d ( ) i \u2022 T \u662f\u4e00\u500b K \u6b21\u591a\u9805\u5f0f\uff0c\u5247(\u5f0f 4-8)\u4e2d\u4ee5\u8655\uf9e4 ( ) i \u2022 T \uf92d\u6700\u5c0f\u5316 i J \uff0c\u5c31\u8b8a\u6210 \u4e00\u500b\u5178\u578b\u7684\u6700\u5c0f\u5316\u5e73\u65b9(least squares)\u7684\u554f\u984c\uff0c\u5982(\u5f0f 4-9)\uff1a ( ) ( ) ( ) ( ) 1 1 0 i K i K i i K K u a u a u a \u2212 \u2212 = + + + T (\u5f0f 4-9) (\u5f0f 4-8)\u4e2d\u7684\u76ee\u6a19\u51fd\uf969\u53ef\u4ee5\u6539\u5beb\u6210\u5411\uf97e\u77e9\u9663\u7684\u5f62\u5f0f\uff0c\u5982(\u5f0f 4-10)\uff1a 2 = i i i i J Y a -b (\u5f0f 4-10) \u5176\u4e2d\u77e9\u9663 i Y \u7684\u7b2c(m,n)\u9805\u5982(\u5f0f 4-11)\u6240\u793a\uff1a ( ) ( ) 1 [ [ ] ] ,1 ,1 1 K n i mn i m m N P n K \u2212 + = \u2264 \u2264 \u2264 \u2264 + Y y (\u5f0f 4-11) \u4e14 ( ) ( ) ( ) 1 0 [ ] i i i T i K K a a a \u2212 = a \uff0c ( ) ( ) ( ) 1 2 T i i i i NP P P P \u23a1 \u23a4 \u23a1 \u23a4 \u23a1 \u23a4 \u23a1 \u23a4 \u23a1 \u23a4 \u23a1 \u23a4 \u23a1 \u23a4 = \u23a2 \u23a5 \u23a2 \u23a5 \u23a2 \u23a5 \u23a2 \u23a5 \u23a2 \u23a5 \u23a2 \u23a5 \u23a2 \u23a5 \u23a2 \u23a5 \u23a2 \u23a5 \u23a2 \u23a5 \u23a3 \u23a6 \u23a3 \u23a6 \u23a3 \u23a6 \u23a3 \u23a6 b x x x \u3002 \u591a\u9805\u5f0f ( ) i \u2022 T \u4e2d\u6700\u5c0f\u5316 i J \u7684\u4fc2\uf969\u5411\uf97e i a \u5373\u70ba\u6700\u5c0f\u5e73\u65b9\u89e3\uff0c\u5982\u4e0b(\u5f0f 4-12)\uff1a ( ) 1 T T i i i i i \u2212 = a Y Y Y b (\u5f0f 4-12) \u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0c\u591a\u9805\u5f0f ( ) i \u2022 T \u7684\u6b21\uf969 K \uf967\u53ef\u4ee5\u8a2d\u592a\u5927\uff0c\u4ee5\u907f\u514d\u6709\u904e\ufa01\u64ec\u5408(over-fitting)\u60c5\u6cc1\u6216\uf967 \uf97c\uf9fa\u6cc1\u7684\u77e9\u9663(ill-conditional matrix) T i i Y Y \u7522\u751f\u3002\u56e0\u6b64\uff0c\u6211\u5011\u53ea\u8003\u616e K = 1 \u8207 K = 2 \uf978\u7a2e\u60c5\u6cc1\uff1a\u7576 K = 1 \u6642\uff0c\u8f49\u79fb\u51fd\uf969 ( ) i \u2022 T \u662f\u4e00\u500b\u7dda\u6027\u51fd\uf969\uff0c\u6211\u5011\u7a31\u4e4b\u70ba\u7dda\u6027\u6700\u5c0f\u5e73\u65b9\u56de\u6b78\u6cd5(", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "Multiband and adaptation approaches to robust speech recognition", |
| "authors": [ |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Tiberewala", |
| "suffix": "" |
| }, |
| { |
| "first": "H", |
| "middle": [], |
| "last": "Hermansky", |
| "suffix": "" |
| } |
| ], |
| "year": 1997, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "107--110", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "S. Tiberewala and H. Hermansky, \"Multiband and adaptation approaches to robust speech recognition\", Eurospeech97, 1997, pp. 107-110", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "Noise robust HMM-based speech recognition using segmental cepstral feature vector normalization", |
| "authors": [ |
| { |
| "first": "O", |
| "middle": [], |
| "last": "Viikki", |
| "suffix": "" |
| }, |
| { |
| "first": "K", |
| "middle": [], |
| "last": "Laurila", |
| "suffix": "" |
| } |
| ], |
| "year": 1997, |
| "venue": "ESCA NATO Workshop Robust Speech Recognition Unknown Communication Channels", |
| "volume": "", |
| "issue": "", |
| "pages": "107--110", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "O. Viikki and K. Laurila, \"Noise robust HMM-based speech recognition using segmental cepstral feature vector normalization\", in ESCA NATO Workshop Robust Speech Recognition Unknown Communication Channels, Pont-a-Mousson, France, 1997, pp.107-110.", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "Feature extraction from higher-lag autocorrelation coefficients for robust speech recognition", |
| "authors": [ |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Benjamin", |
| "suffix": "" |
| }, |
| { |
| "first": "Kuldip", |
| "middle": [ |
| "K" |
| ], |
| "last": "Shannon", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Paliwal", |
| "suffix": "" |
| } |
| ], |
| "year": 2006, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Benjamin J. Shannon, Kuldip K. Paliwal, \"Feature extraction from higher-lag autocorrelation coefficients for robust speech recognition\", Speech Communication 2006.", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "Effectiveness of linear prediction characteristics of the speech wave for automatic speaker identification and verification", |
| "authors": [ |
| { |
| "first": "B", |
| "middle": [ |
| "S" |
| ], |
| "last": "Atal", |
| "suffix": "" |
| } |
| ], |
| "year": 1974, |
| "venue": "Journal of the Acoustical Society of America", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Atal, B.S. \"Effectiveness of linear prediction characteristics of the speech wave for automatic speaker identification and verification\", Journal of the Acoustical Society of America, 1974.", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "Spectral linear prediction: properties and applications", |
| "authors": [ |
| { |
| "first": "J", |
| "middle": [], |
| "last": "", |
| "suffix": "" |
| } |
| ], |
| "year": 1975, |
| "venue": "IEEE Transactions on Acoustics, Speech and Signal Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "J. Makhoul, \"Spectral linear prediction: properties and applications,\" IEEE Transactions on Acoustics, Speech and Signal Processing, 1975.", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "Perceptual linear predictive (PLP) analysis of speech", |
| "authors": [ |
| { |
| "first": "H", |
| "middle": [], |
| "last": "Hermansky", |
| "suffix": "" |
| } |
| ], |
| "year": 1990, |
| "venue": "J. Acoust. Soc. Am", |
| "volume": "87", |
| "issue": "4", |
| "pages": "1738--1752", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "H. Hermansky, \"Perceptual linear predictive (PLP) analysis of speech\", J. Acoust. Soc. Am, vol. 87, no. 4, pp. 1738-1752, Apr. 1990.", |
| "links": null |
| }, |
| "BIBREF6": { |
| "ref_id": "b6", |
| "title": "Cepstral statistics compensation using online pseudo stereo codebooks for robust speech recognition in additive noise environments", |
| "authors": [ |
| { |
| "first": "Jeih-Weih", |
| "middle": [], |
| "last": "Hung", |
| "suffix": "" |
| } |
| ], |
| "year": 2006, |
| "venue": "ICASSP", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Jeih-weih Hung, \"Cepstral statistics compensation using online pseudo stereo codebooks for robust speech recognition in additive noise environments\", ICASSP 2006.", |
| "links": null |
| }, |
| "BIBREF7": { |
| "ref_id": "b7", |
| "title": "RASTA processing of speech", |
| "authors": [ |
| { |
| "first": "H", |
| "middle": [], |
| "last": "Hermansky", |
| "suffix": "" |
| }, |
| { |
| "first": "N", |
| "middle": [], |
| "last": "Morgan", |
| "suffix": "" |
| } |
| ], |
| "year": 1994, |
| "venue": "IEEE Transactions on Speech and Audio Processing", |
| "volume": "2", |
| "issue": "", |
| "pages": "578--589", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "H. Hermansky and N. Morgan, \"RASTA processing of speech\", IEEE Transactions on Speech and Audio Processing, 2, pp.578-589, 1994", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "FIGREF0": { |
| "num": null, |
| "text": "\u7684 \u6574 \u6bb5 \u5f0f \u5012 \u983b \u8b5c \u5e73 \u5747 \u8207 \u8b8a \uf962 \uf969 \u6b63 \u898f \u5316 \u6cd5 [1](utterance-based cepstral mean and variance normalization\uff0cU-CMVN)\u8207\u5206\u6bb5\u5f0f\u5012\u983b\u8b5c\u5e73\u5747\u8207\u8b8a\uf962\uf969\u6b63\u898f\u5316\u6cd5[2](segmental cepstral mean and variance normalization\uff0cS-CMVN)\u3002\u524d\u8005\u662f\u4ee5\u4e00\u6574\u6bb5\u8a9e\uf906\u70ba\u57fa\u6e96\u53bb\u4f30\u7b97\u8a72\u7dad\u7279\u5fb5\uf96b\uf969\u7684 statistics compensation, CSC)\u3001\u7dda\u6027\u6700\u5c0f\u5e73\u65b9\u56de\u6b78\u6cd5(linear least squares regression, LLS)\u8207\u4e8c\u6b21\u6700\u5c0f\u5e73\u65b9\u56de\u6b78\u6cd5(quadratic least squares regression, QLS)\u3002 -frequency cepstral coefficients\uff0cMFCC)\u3001\u81ea\u76f8\u95dc\u6885\u723e\u5012\u983b\u8b5c\u4fc2\uf969[3](autocorrelation mel-frequency cepstral coefficients\uff0cAMFCC )\u3001\u7dda\u6027\u9810\u6e2c\u5012\u983b\u8b5c\u4fc2\uf969[4][5](linear prediction cepstral coefficients \uff0c LPCC) \u4ee5 \u53ca \u611f \u77e5 \u7dda \u6027 \u9810 \u6e2c \u5012 \u983b \u8b5c \u4fc2 \uf969 [6](perceptual linear prediction cepstral coefficients\uff0cPLPCC)\u3002 \u6211\u5011\u5c07\u4f7f\u7528\u9019\u56db\u7a2e\u8a9e\u97f3\u7279\u5fb5\uf96b\uf969\uf92d\u9a57\u8b49\u672c\uf941\u6587\u6240\u63d0\u51fa\u7684\u5f37\u5065\u6027\u8a9e\u97f3\u7279 \u5fb5\uf96b\uf969\u6280\u8853\uff0c\u4e26\u4e14\u8207\u5176\u4ed6\u5f37\u5065\u6027\u65b9\u6cd5\u904b\u7528\u5728\u9019\u56db\u7a2e\u7279\u5fb5\uf96b\uf969\u4e0a\u505a\u6bd4\u8f03\u3002 (\u4e00) \u6885\u723e\u5012\u983b\u8b5c\u4fc2\uf969 (mel-frequency cepstral coefficients\uff0cMFCC)", |
| "type_str": "figure", |
| "uris": null |
| }, |
| "FIGREF1": { |
| "num": null, |
| "text": "\u7dda\u6027\u9810\u6e2c\u5012\u983b\u8b5c\u4fc2\uf969(linear prediction cepstral coefficients\uff0cLPCC) \u7dda\u6027\u9810\u6e2c(linear prediction)\u7684\u57fa\u672c\u539f\uf9e4\u662f\u5047\u8a2d\u76ee\u524d\u7684\u8072\u97f3\u53d6\u6a23\u503c\u53ef\u7531\u5728\u524d\u9762\u7684 p \u500b\u53d6\u6a23 \u503c\uff0c\u4ee5\u7dda\u6027\u7d44\u5408\uf92d\u9810\u6e2c\u3002\u5716\u4e09\u5373\u70ba\u7dda\u6027\u9810\u6e2c\u5012\u983b\u8b5c\u4fc2\uf969\u4e4b\u64f7\u53d6\uf9ca\u7a0b\u5716\u3002\u5982\u540c\u524d\uf978\u7a2e\u7279\u5fb5\uf96b\uf969\u64f7 \u53d6\u6280\u8853\uff0c\u6211\u5011\u5c07\u8a9e\u97f3\u8a0a\u865f\u7d93\u904e\u9810\u5f37\u8abf\u5f8c\uff0c\ufa00\u5272\u6210\u8a31\u591a\u4e00\u5c0f\u6bb5\u7684\u97f3\u6846\u8207\u6f22\u660e\u8996\u7a97\u7684\u8655\uf9e4\u5f8c\u53d6\u5176\u81ea \u76f8\u95dc\u4fc2\uf969\uff0c\u900f\u904e Levinson Durbin \u6f14\u7b97\u6cd5\u6c42\u5f97\u7dda\u6027\u9810\u6e2c\u4fc2\uf969\uff0c\u6700\u5f8c\u5c07\u7dda\u6027\u9810\u6e2c\u4fc2\uf969\u8f49\u63db\u6210\u5012\u983b \u8b5c\uff0c\uf965\u5f97\u5230\u7dda\u6027\u9810\u6e2c\u5012\u983b\u8b5c\u4fc2\uf969(linear prediction cepstral coefficients\uff0cLPCC)\u3002", |
| "type_str": "figure", |
| "uris": null |
| }, |
| "FIGREF2": { |
| "num": null, |
| "text": "\u6574 \u6bb5 \u5f0f \u5012 \u983b \u8b5c \u5e73 \u5747 \u503c \u8207 \u8b8a \uf962 \uf969 \u6b63 \u898f \u5316 \u6cd5 (utterance-based cepstral mean and variance normalization\uff0cU-CMVN)", |
| "type_str": "figure", |
| "uris": null |
| }, |
| "FIGREF3": { |
| "num": null, |
| "text": "-line manner)\uff0c\u5373\u5728\uf967\u6703\u6709\u592a\u9577\u7684\u5ef6\u9072\u6642\u9593\u7684\u904b\u7b97\u60c5\u6cc1\u4e0b\u5efa\uf9f7\u3002 \u5728\u672c\uf941\u6587\u4e2d\uff0c\u6211\u5011\u4ee5\u865b\u64ec\u96d9\u901a\u9053\u78bc\u7c3f\u70ba\u57fa\u790e\uf92d\u57f7\ufa08\u4e09\u7a2e\u7279\u5fb5\uf96b\uf969\u88dc\u511f\u6cd5\uff0c\u4ee5\ufa09\u4f4e\u52a0\u6210\u6027\u96dc \u8a0a\u7684\u5f71\u97ff\u3002\u4ee5\u4e0b\uff0c\u6211\u5011\u5c0d\u4e09\u7a2e\u7279\u5fb5\uf96b\uf969\u88dc\u511f\u6cd5\u505a\u5b8c\u6574\u7684\u4ecb\u7d39\u3002 (\u4e8c) \u5012\u983b\u8b5c\u7d71\u8a08\u88dc\u511f\u6cd5(cepstral statistics compensation\uff0cCSC)", |
| "type_str": "figure", |
| "uris": null |
| }, |
| "TABREF5": { |
| "text": "Markov model toolkit\uff0cHTK)\uf92d\u8a13\uf996\uff0c\u5305\u542b 11 \u500b\uf969\u5b57\u6a21\u578b(0~9 \u4ee5\u53ca oh 11 \u500b\uf969\u5b57\u6a21\u578b)\u4ee5\u53ca\u975c\u97f3\u6a21\u578b\uff0c\u6bcf\u500b\uf969\u5b57\u6a21\u578b\u5305\u542b 10 \u500b\uf9fa\u614b\uff0c\u5404\uf9fa\u614b\u5305\u542b 4 \u500b\u9ad8\u65af\u5bc6\ufa01\u6df7\u5408\u3002\u96b1 \u503c\u90fd\uf967\u76f8\u540c\uff0c\uf967\u904e\uf974\u500b\u5225\u89c0\u5bdf\u500b\u7279\u5fb5\uf96b\uf969\u7684\u5be6\u9a57\u7d50\u679c\uff0c\u53ef\u767c\u73fe\u5176\u5177\u6709\u898f\u5247\u6027\u3002\u5982\u5728\u7279\u5fb5\uf96b \uf969\u70ba MFCC \u6642\uff0c\u4e09\u7a2e\u7279\u5fb5\uf96b\uf969\u88dc\u511f\u6cd5\u4e4b\u6700\u4f73\u5be6\u9a57\u7d50\u679c\u90fd\u5728 N \u503c\u70ba 512 \u6216 256 \u9019\u4e9b\u6bd4\u8f03\u4e2d\u6bb5 \u7684\u503c\uff1b\u800c\u7279\u5fb5\uf96b\uf969\u70ba LPCC \u6642\u5247\u5728 N \u503c\u70ba\u8f03\u5927\u503c 1024 \u6642\uff0c\u53ef\u5f97\u5230\u6700\u4f73\u8fa8\uf9fc\uf961\u3002", |
| "num": null, |
| "content": "<table><tr><td>(AMFCC)\u3001\u7dda\u6027\u9810\u6e2c\u5012\u983b\u8b5c\u4fc2\uf969(LPCC)\u53ca\u611f\u77e5\u7dda\u6027\u9810\u6e2c\u5012\u983b\u8b5c\u4fc2\uf969(PLPCC)\uff0c\u5176\u76f8\u95dc\u8a2d\u5b9a\u8207\u500b \u5e73\u5747\u8207\u8b8a\uf962\uf969\u6b63\u898f\u5316\u6cd5(S-CMVN)\u7684\u8fa8\uf9fc\u7cbe\u78ba\ufa01\uff0c\u76f8\u5c0d\u65bc\u539f\u59cb\u672a\u8655\uf9e4\u7684\u5404\u7a2e\u5012\u983b\u8b5c\u7279\u5fb5\u800c\u8a00\uff0c \u8868\u4e03\u3001\u4e8c\u6b21\u6700\u5c0f\u5e73\u65b9\u56de\u6b78\u6cd5(QLS)\u4e4b\u8fa8\uf9fc\u7cbe\u78ba\ufa01(%)\uff0c\u5176\u4e2d N \u8868\u793a\u4e7e\u6de8\u78bc\u7c3f\u7684\u78bc\u5b57\uf969</td></tr><tr><td>\u5225\u5c0d\u61c9\u4e4b\u4e2d\u4ecb\u7279\u5fb5\uf96b\uf969\uff0c\u5982\u8868\u4e8c\u6240\u793a\u3002\u5c0d\u65bc\u6bcf\u500b\u6b32\u8fa8\uf9fc\u7684\uf969\u5b57\u6a21\u578b\u800c\u8a00\uff0c\u672c\uf941\u6587\u4f7f\u7528\u96b1\u85cf\u5f0f\u99ac U-CMVN \u8207 S-CMVN \u7686\u80fd\u6709\u6548\u63d0\u5347\u5404\u7a2e\u96dc\u8a0a\u74b0\u5883\u4e0b\u7684\u8fa8\uf9fc\uf961\uff0c\u9019\u610f\u8b02\u9019\uf978\u7a2e\u65b9\u6cd5\u7684\u78ba\u5177\u6709\u63d0 4. \u4e00\u822c\u800c\u8a00\uff0c\u5012\u983b\u8b5c\u7d71\u8a08\u88dc\u511f\u6cd5\u7684\u6548\u679c\u512a\u65bc\u7dda\u6027\u6700\u5c0f\u5e73\u65b9\u56de\u6b78\u6cd5\u8207\u4e8c\u6b21\u6700\u5c0f\u5e73\u65b9\u56de\u6b78\u6cd5\uff0c\u7136 S-CMVN average exhibition car babble subway Set A S-CMVN average exhibition car babble subway Set A</td></tr><tr><td>\u53ef\u592b\u6a21\u578b\u5de5\u5177(hidden \u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b\u662f\u4e00\u7a2e\u904b\u7528\u7d71\u8a08\uf9e4\uf941\u63a8\u5c0e\u51fa\uf92d\u7684\u6a21\u578b\uff0c\u7528\uf92d\u63cf\u8ff0\u8a9e\u97f3\u7522\u751f\u7684\u904e\u7a0b\uff0c\u76f8\u7576\u9069\u5408\u7528 \u6e2c\u7279\u5fb5\uf96b\uf969\u7684\u7d71\u8a08\u503c\u80fd\u6bd4\uf9dd\u7528\u6574\u6bb5\u7684\u65b9\u5f0f\uf901\u7cbe\u78ba\u3002 \u8868\u4e94\u3001\u5012\u983b\u8b5c\u7d71\u8a08\u88dc\u511f\u6cd5(CSC)\u4e4b\u8fa8\uf9fc\u7cbe\u78ba\ufa01(%)\uff0c\u5176\u4e2d N \u8868\u793a\u4e7e\u6de8\u78bc\u7c3f\u7684\u78bc\u5b57\uf969 74.60 72.54* 69.54 77.81 75.45 67.35 AC-LPCC (N=1024) 74.16 71.04 80.80 74.99 69.82 MS-LPCC (N=1024) 72.54* 69.54 77.81 75.45 67.35 AC-LPCC (N=1024) 74.60 74.16 71.04 80.80 74.99 69.82 MS-LPCC (N=1024) U-CMVN \u5728\u8fa8\uf9fc\uf961\u80fd\u6709\uf901\u660e\u986f\u7684\u63d0\u5347\u3002\u9019\u8207\u6211\u5011\u4e4b\u524d\u5206\u6790\u7684\u7d50\u679c\u76f8\u543b\u5408\uff0c\u5373\uf9dd\u7528\u5206\u6bb5\u7684\u65b9\u5f0f\u4f30 72.66 75.41 71.02 81.08 76.94 72.61 AMFCC (N=512) 72.66 75.41 71.02 81.08 76.94 AMFCC (N=512) 72.61 \u5347\u7279\u5fb5\uf96b\uf969\u5f37\u5065\u6027\u7684\u6548\u80fd\uff0c\u800c\u7576\u6211\u5011\u5c07\u8868\u56db\u8207\u8868\u4e09\u7684\uf969\u64da\u6bd4\u8f03\uff0c\u53ef\u660e\u986f\u770b\u51fa S-CMVN \u76f8\u5c0d\u65bc \u800c\uff0c\u5176\u8868\u73fe\u7684\u5dee\uf962\u4e26\u6c92\u6709\u5341\u5206\u660e\u986f\u3002 73.53 78.53 77.38 82.60 76.44 77.71 MFCC (N=256) 73.53 78.53 77.38 82.60 76.44 77.71 MFCC (N=256)</td></tr><tr><td>\u5728\uf99a\u7e8c\u8a9e\u97f3\u7684\u8fa8\u8a8d\u3002HMM \u6709\u5f88\u591a\u7a2e\uf9d0\u578b\uff0c\u672c\uf941\u6587\u63a1\u7528\u7531\u5de6\u5230\u53f3\u7684\u5f62\u5f0f\uff0c\u4e5f\u5c31\u662f\u6bcf\u500b\uf9fa\u614b\u5728\u4e0b \u4e00\u500b\u6642\u9593\u53ea\u80fd\u8df3\u5230\u6b64\u523b\uf9fa\u614b\u6216\u4e0b\u4e00\u500b\u9130\u8fd1\u7684\uf9fa\u614b\uff0c\u96a8\u8457\u6642\u9593\u7684\u589e\u52a0\uff0c\uf9fa\u614b\u7531\u5de6\u81f3\u53f3\u4f9d\u5e8f\u8f49\u79fb\u3002 \u63a5\u4e0b\uf92d\uff0c\u6211\u5011\u63a2\u8a0e\u672c\uf941\u6587\u6240\u63d0\u51fa\u7684\u4e09\u7a2e\u4ee5\u78bc\u7c3f\u70ba\u57fa\u790e\u7684\u7279\u5fb5\uf96b\uf969\u88dc\u511f\u6cd5\u7684\u6548\u679c\uff0c\u8868\u4e94\u3001\u8868\uf9d1\u8207 73.53 78.12 77.40 80.54 75.84 78.71 MFCC (N=512) 73.53 78.12 77.40 80.54 75.84 78.71 MFCC (N=512) S-CMVN average train station airport street restaurant Set B S-CMVN average train station airport street restaurant Set B \u8868\u4e03\u5206\u5225\u70ba\u5012\u983b\u8b5c\u7d71\u8a08\u88dc\u511f\u6cd5(CSC)\u3001\u7dda\u6027\u6700\u5c0f\u5e73\u65b9\u56de\u6b78\u6cd5(LLS)\u8207\u4e8c\u6b21\u6700\u5c0f\u5e73\u65b9\u56de\u6b78\u6cd5(QLS) S-CMVN average exhibition car babble subway Set A S-CMVN average exhibition car babble subway Set A 75.75 78.58 77.94 82.80 76.74 76.83 MS-PLPCC (N=64) 72.82* 72.48 74.29 73.05 71.48 AC-PLPCC (N=512) 78.58 77.94 82.80 76.74 76.83 MS-PLPCC (N=64) 75.75 72.82* 72.48 74.29 73.05 71.48 AC-PLPCC (N=512)</td></tr><tr><td>\u53e6\u5916\uff0c\u6a21\u578b\u4e2d\u7684\uf9fa\u614b\u89c0\u6e2c\u6a5f\uf961\u51fd\uf969\u662f\u9078\u7528\uf99a\u7e8c\u5f0f\u7684\u9ad8\u65af\u6df7\u5408\u6a5f\uf961\u5bc6\ufa01\u51fd\uf969(Gaussian Mixture \u7684\u8fa8\uf9fc\u7cbe\u78ba\ufa01\u3002\u70ba\uf9ba\u6bd4\u8f03\u8d77\ufa0a\uff0c\u6211\u5011\u5c07\u8868\u56db\u4e4b S-CMVN \u7684\u7d50\u679c\u4ea6\uf99c\u65bc\u5404\u8868\u4e2d\u3002\u5f9e\u9019\u4e09\u500b\u8868\u7684\uf969 72.66 78.02 75.91 82.80 74.26 79.11 AMFCC (N=512) 72.66 78.02 75.91 82.80 74.26 79.11 AMFCC (N=512) 74.45 76.99 79.18 77.70 77.19 73.88 MFCC (N=256) 74.45 76.99 79.18 77.70 77.19 73.88 MFCC (N=256)</td></tr><tr><td>probability density function\uff0c\u7c21\u7a31 GM)\uff0c\u56e0\u6b64\u6211\u5011\u4e5f\u7a31\u6b64\u6a21\u578b\u70ba\uf99a\u7e8c\u5bc6\ufa01\u96b1\u85cf\u5f0f\u99ac\u53ef\u592b\u6a21\u578b (continuous density HMM\uff0c\u7c21\u7a31 CDHMM)\u3002 \u8868\u4e8c\u3001\u5be6\u9a57\u4e2d\u6240\u7528\u7684\u7279\u5fb5\uf96b\uf969\u8a73\u7d30\u8cc7\uf9be \u7279\u5fb5\uf96b\uf969\u7a2e\uf9d0 \u7279\u5fb5\uf96b\uf969\u7dad\ufa01 \u64da\u53ef\u77e5\uff1a \u8868\u4e09\u3001\u6574\u6bb5\u5f0f\u5012\u983b\u8b5c\u5e73\u5747\u8207\u8b8a\uf962\uf969\u6b63\u898f\u5316\u6cd5\u4e4b\u8fa8\uf9fc\u7cbe\u78ba\ufa01(%) MFCC Set A 72.91 subway 69.71 babble 68.71 car 69.22 exhibition 61.99 70.14 baseline average MFCC Set A 72.91 subway 69.71 babble 68.71 car 69.22 exhibition 61.99 70.14 baseline average S-CMVN 75.75 74.60 78.47 78.71 81.25 76.14 77.79 MS-PLPCC (N=128) 76.60 74.54 81.19 75.71 74.94 AC-LPCC (N=1024) average train station airport street restaurant Set B 77.87 77.70 78.55 76.57 78.64 AC-PLPCC (N=32) 76.28 74.62 82.38 72.31 75.81 MS-LPCC (N=1024) S-CMVN 76.28 74.62 82.38 72.31 75.81 MS-LPCC (N=1024) 74.71 76.91 78.72 78.95 75.50 74.47 AMFCC(N=512) 74.71 76.91 78.72 78.95 75.50 74.47 AMFCC(N=512) 74.60 78.47 78.71 81.25 76.14 77.79 MS-PLPCC (N=128) 76.60 74.54 81.19 75.71 74.94 AC-LPCC (N=1024) 75.86 75.34* 76.97 78.92 70.17 75.28 AC-LPCC (N=1024) 76.25 78.55 79.65 72.06 74.75 MS-LPCC (N=1024) 76.25 78.55 79.65 72.06 74.75 MS-LPCC (N=1024) 75.86 75.34* 76.97 78.92 70.17 75.28 AC-LPCC (N=1024) 75.75 average train station airport street restaurant Set B 77.87 77.70 78.55 76.57 78.64 AC-PLPCC (N=32) 77.13 78.89 81.19 81.22 76.60 76.53 MS-PLPCC (N=64) 73.41* 74.24 76.19 70.62 72.61 AC-PLPCC (N=512) 78.89 81.19 81.22 76.60 76.53 MS-PLPCC (N=64) 77.13 73.41* 74.24 76.19 70.62 72.61 AC-PLPCC (N=512) \u4e2d\u4ecb\u7279\u5fb5\uf96b\uf969\u7dad\ufa01 MFCC 12 \u7dad\u5012\u983b\u8b5c\u52a0\u4e0a 1 \u5c0d\uf969\u80fd\uf97e \u7dad\uff0c\u4e26\u53d6\u5176\u4e00\u968e\u548c\u4e8c\u968e\u5dee\uf97e\uff0c 23 \u7dad\u6885\u723e\u983b\u8b5c\u52a0\u4e0a 1 \u5c0d\uf969\u80fd\uf97e\u7dad\u3002 PLPCC LPCC AMFCC 75.24 72.25 68.60 74.34 71.32 72.02 73.72 71.50 68.94 74.60 70.18 65.58 57.38 74.48 51.26 71.31 65.52 68.78 PLPCC LPCC AMFCC 75.24 72.25 68.60 74.34 71.32 72.02 73.72 71.50 68.94 74.60 70.18 65.58 57.38 74.48 51.26 71.31 65.52 68.78 75.86 74.71 74.45 78.04 79.61 80.24 75.95 76.35 AC-LPCC (N=1024) 77.22 79.10 77.28 79.36 73.15 AMFCC (N=512) 76.95 77.77 77.15 77.82 75.08 MFCC (N=512) 76.93 79.16 78.48 76.55 73.57 MS-LPCC (N=1024) 74.71 74.45 77.22 79.10 77.28 79.36 73.15 76.95 77.77 77.15 77.82 75.08 MFCC (N=512) \u4ee5\u7dda\u4e0a\u65b9\u5f0f\u5373\u6642\u5730\u4f30\u7b97\u51fa\u96dc\u8a0a\u8a9e\u97f3\u7279\u5fb5\uf96b\uf969\u7d71\u8a08\u503c\uff0c\u6240\u4f30\u7b97\u51fa\u7684\u7d71\u8a08\u503c\u8f03\u70ba\u6e96\u78ba\uff0c\u4e5f\u4f7f\u5f97\u57f7\ufa08 AMFCC (N=512) 76.93 79.16 78.48 76.55 73.57 \u7279\u5fb5\uf96b\uf969\u88dc\u511f\u6cd5\u5f8c\u8a9e\u97f3\u7279\u5fb5\uf96b\uf969\uf901\u70ba\u5f37\u5065\u3002\u76f8\u5c0d\u65bc\u50b3\u7d71\u7279\u5fb5\uf96b\uf969\u6b63\u898f\u5316\u6cd5\u662f\u4ee5\u6574\u6bb5\u6216\u5206\u6bb5\u8a9e\uf906 MS-LPCC (N=1024) 75.86 78.04 79.61 80.24 75.95 76.35 AC-LPCC (N=1024) \u70ba\u57fa\u790e\u53bb\u4f30\u7b97\u8a9e\u97f3\u7279\u5fb5\uf96b\uf969\u7684\u7d71\u8a08\u503c\u5f8c\uff0c\u800c\u57f7\ufa08\u7279\u5fb5\uf96b\uf969\u6b63\u898f\u5316\uff0c\u4ee5\u865b\u64ec\u96d9\u901a\u9053\u78bc\u7c3f\u70ba\u57fa\u790e\u7684 \u7e3d\u5171 39 \u7dad\u7279\u5fb5\uf96b\uf969\u3002 AMFCC 12 \u7dad\u5012\u983b\u8b5c\u52a0\u4e0a 1 \u5c0d\uf969\u80fd\uf97e \u7dad\uff0c\u4e26\u53d6\u5176\u4e00\u968e\u548c\u4e8c\u968e\u5dee\uf97e\uff0c AMFCC MFCC Set B 72.89 71.60 restaurant 71.23 72.16 street 73.35 71.00 airport 70.23 68.28 train station 59.43 71.93 55.78 70.76 baseline average AMFCC MFCC Set B 72.89 71.60 restaurant 71.23 72.16 street 73.35 71.00 airport 70.23 68.28 train station 59.43 55.78 70.76 baseline average 77.13 78.66 79.55 80.40 77.47 77.24 MS-PLPCC (N=128) 77.95 78.07 79.90 77.06 76.76 AC-PLPCC (N=32) 78.66 79.55 80.40 77.47 77.24 MS-PLPCC (N=128) \u7279\u5fb5\uf96b\uf969\u88dc\u511f\u6cd5\uff0c\uf901\u80fd\ufa09\u4f4e\u96dc\u8a0a\u5c0d\u8a9e\u97f3\u7684\u5f71\u97ff\u3002 77.13 77.95 78.07 79.90 77.06 76.76 AC-PLPCC (N=32) \u672c\uf941\u6587\u53ea\u8457\u91cd\u65bc\u52a0\u6210\u6027\u96dc\u8a0a\u74b0\u5883\u4e0b\u7684\u7814\u7a76\uff0c\u56e0\u6b64\u5728\u672a\uf92d\uff0c\u6211\u5011\u671f\u671b\u80fd\u4ee5\u865b\u64ec\u78bc\u7c3f\u70ba\u57fa\u790e\u7684 71.93 23 \u7dad\u6885\u723e\u983b\u8b5c\u52a0\u4e0a 1 \u5c0d\uf969\u80fd\uf97e\u7dad\u3002 LPCC 73.44 72.82 74.68 70.69 49.58 72.91 LPCC 73.44 72.82 74.68 70.69 49.58 72.91 \u5f37\u5065\u6027\u8a9e\u97f3\u6280\u8853\uff0c\u85c9\u7531\u7d50\u5408\u4e00\u4e9b\u901a\u9053\u88dc\u511f\u6280\u5de7\u5982\uff1a\u76f8\u5c0d\u983b\u8b5c\u6cd5(RASTA) [8]\uff0c\u4f7f\u9019\u4e9b\u865b\u64ec\u78bc\u7c3f\u70ba</td></tr><tr><td>\u7e3d\u5171 39 \u7dad\u7279\u5fb5\uf96b\uf969\u3002 76.48 75.79 76.48 75.79 \u8868\uf9d1\u3001\u7dda\u6027\u6700\u5c0f\u5e73\u65b9\u56de\u6b78\u6cd5(LLS)\u4e4b\u8fa8\uf9fc\u7cbe\u78ba\ufa01(%)\uff0c\u5176\u4e2d N \u8868\u793a\u4e7e\u6de8\u78bc\u7c3f\u7684\u78bc\u5b57\uf969 PLPCC 77.01 73.23 54.51 75.63 PLPCC 77.01 73.23 54.51 75.63 \u57fa\u790e\u7684\u5f37\u5065\u6027\u8a9e\u97f3\u6280\u8853\u80fd\u5ef6\u4f38\u65bc\u6d88\u9664\u901a\u9053\u5931\u771f\u7684\u6548\u61c9\u4e0a\u3002</td></tr><tr><td>linear least squares \u5373\u70ba\u4e00\u500b\u4e8c\u6b21\u51fd\uf969\uff0c\u6211\u5011\u7a31\u4e4b\u70ba\u4e8c\u6b21\u6700\u5c0f\u5e73\u65b9\u56de\u6b78 13 \u7dad\u5012\u983b\u8b5c\uff0c\u4e26\u53d6\u5176\u4e00\u968e\u548c\u4e8c regression, LLS)\u3002\u7576 K = 2 \u6642\u8f49\u79fb\u51fd\uf969 ( ) i \u2022 T 23 \u7dad\u5f37\ufa01\u983b\u8b5c\uff0c \u8868\u56db\u3001\u5206\u6bb5\u5f0f\u5012\u983b\u8b5c\u5e73\u5747\u8207\u8b8a\uf962\uf969\u6b63\u898f\u5316\u6cd5\u4e4b\u8fa8\uf9fc\u7cbe\u78ba\ufa01(%) S-CMVN average exhibition car babble subway Set A S-CMVN average exhibition car babble subway Set A LPCC \u968e\u5dee\uf97e\uff0c\u7e3d\u5171 39 \u7dad\u7279\u5fb5\uf96b\uf969\u3002 \u6216\u662f 24 \u7dad\u81ea\u76f8\u95dc\u4fc2\uf969\u3002 U-CMVN subway babble car exhibition average U-CMVN subway babble car exhibition average 73.53 77.90 76.57 80.01 76.09 78.92 MFCC (N=512) 73.53 77.90 76.57 80.01 76.09 78.92 MFCC (N=512) \u516b\u3001\uf96b\u8003\u6587\u737b</td></tr><tr><td>\u6cd5(quadratic least squares regression, QLS)\u3002 \u5982\u540c\u672c\u7bc0\u4e00\u958b\u59cb\u6240\u63d0\u5230\uff0c\u7528\u9019\uf978\u7a2e\u591a\u9805\u5f0f\u56de\u6b78\u6cd5\u7684\u6982\uf9a3\u5c31\u662f\u5e0c\u671b\u96dc\u8a0a\u8a9e\u97f3\u7684\u78bc\u7c3f\uff0c\u5728\u900f\u904e \u4e00\u500b\u8f49\u63db\u51fd\uf969\u7684\u904b\u7b97\u5f8c\u80fd\u548c\u4e7e\u6de8\u8a9e\u97f3\u78bc\u7c3f\u7684\u6574\u9ad4\u8ddd\uf9ea\u662f\u6700\u5c0f\u7684\uff0c\u7576\u96dc\u8a0a\u8a9e\u97f3\u5012\u983b\u8b5c\u7d93\u904e\u76f8\u540c\u8f49 \u63db\u5f8c\u6703\uf901\u63a5\u8fd1\u4e7e\u6de8\u8a9e\u97f3\u5012\u983b\u8b5c\uff0c\u5982\u6b64\uf965\u53ef\u63d0\u5347\u8fa8\uf9fc\u6548\u679c\u3002 \u968e\u5dee\uf97e\uff0c\u7e3d\u5171 39 \u7dad\u7279\u5fb5\uf96b\uf969\u3002 \u4fc2\uf969\u3002 (\u56db) \u5f37\u5065\u6027\u7279\u5fb5\uf96b\uf969\u6280\u8853\u5be6\u9a57\u8a2d\u5b9a 76.07 74.53 72.12 75.14 74.20 74.82 75.72 75.30 73.49 76.06 74.38 70.21 74.48 75.75 71.31 74.60 68.78 72.66 76.07 74.53 72.12 75.14 74.20 74.82 75.72 75.30 73.49 76.06 74.38 70.21 74.48 75.75 71.31 75.75 79.51 79.22 81.37 77.70 79.74 MS-PLPCC (N=64) 76.43 75.93 76.02 77.15 76.63 AC-PLPCC (N=64) 76.43 75.93 76.02 77.15 76.63 AC-PLPCC (N=64) 75.75 79.51 79.22 81.37 77.70 MS-PLPCC (N=64) 79.74 74.60 68.78 72.66 74.60 74.58 70.91 78.22 77.48 71.70 AC-LPCC (N=1024) 74.77 71.18 79.79 75.23 72.85 MS-LPCC (N=1024) 74.58 70.91 78.22 77.48 71.70 AC-LPCC (N=1024) 74.60 74.77 71.18 79.79 75.23 72.85 MS-LPCC (N=1024) PLPCC 13 \u7dad\u5012\u983b\u8b5c\uff0c\u4e26\u53d6\u5176\u4e00\u968e\u548c\u4e8c 23 \u7dad\u5f37\ufa01\u983b\u8b5c\uff0c\u6216\u662f 23 \u7dad\u81ea\u76f8\u95dc 75.71 73.42 72.36 72.63 70.14 73.53 75.71 73.42 72.36 72.63 70.14 73.53 72.66 76.46 73.46 80.35 75.06 76.97 AMFCC (N=512) 72.66 76.46 73.46 80.35 75.06 76.97 AMFCC (N=512)</td></tr><tr><td>\u5728\u5206\u6bb5\u5f0f\u5e73\u5747\u503c\u8207\u8b8a\uf962\uf969\u6b63\u898f\u5316\u6cd5(S-CMVN)\uff0c\u6211\u5011\uf9a8\u5f0f 3-5 \u8207\u5f0f 3-6 \u4e2d\u4f7f\u7528\u7684\u5206\u6bb5\u9577\ufa01\u70ba restaurant street airport train station U-CMVN average restaurant street airport train station U-CMVN average S-CMVN average train station airport street restaurant Set B S-CMVN average train station airport street restaurant Set B \u4e94\u3001\u5be6\u9a57\u8a2d\u5b9a 75.65 75.23 74.97 71.93 70.76 74.45 75.65 75.23 74.97 71.93 70.76 74.45 74.45 77.35 77.62 77.72 78.19 78.89 MFCC (N=512) 74.45 77.35 77.62 77.72 78.19 78.89 MFCC (N=512) P+1 = 101 \u500b\u97f3\u6846\uff0c\u5373\u5927\u7d04\u70ba 1 \u79d2\u7684\u9577\ufa01\u3002 (\u4e00) \u8a9e\u97f3\u8cc7\uf9be\u5eab\u7c21\u4ecb \u865b\u64ec\u96d9\u901a\u9053\u78bc\u7c3f\u7684\u5efa\uf9f7\u65b9\u6cd5\u7576\u4e2d\uff0c\u4e7e\u6de8\u8a9e\u97f3\u78bc\u7c3f\u70ba { [ ],1 } n n N \u2264 \u2264 x 75.58 76.11 75.02 72.14 71.93 74.71 75.58 76.11 75.02 72.14 71.93 74.71 74.71 76.56 76.95 77.29 77.55 74.45 AMFCC (N=512) 74.71 76.56 76.95 77.29 77.55 74.45 AMFCC (N=512) \uff0c\u5176\u4e2d N \u503c\u6211\u5011\u5206\u5225\u8a2d \u672c\uf941\u6587\u6240\u4f7f\u7528\u7684\u8a9e\u97f3\u8cc7\uf9be\u5eab\u70ba\u6b50\u6d32\u96fb\u4fe1\u6a19\u6e96\u5354\u6703(European Telecommunication Standard Institute\uff0cETSI)\u767c\ufa08\u7684 AURORA2 \u8a9e\u97f3\u8cc7\uf9be\u5eab\uff0c\u5b83\u662f\u4e00\u5957\uf99a\u7e8c\u7684\u82f1\u6587\uf969\u5b57\u5b57\uf905\uff0c\u5167\u5bb9\u662f\u4ee5\u7f8e\u570b \u6210\uf98e\u7537\uf981\u6240\uf93f\u88fd\u7684\u4e7e\u6de8\u74b0\u5883\uf99a\u7e8c\uf969\u5b57\uff0c\u518d\u52a0\u4e0a\u96dc\u8a0a\u8207\u901a\u9053\u6548\u61c9\u3002\u52a0\u6210\u6027\u96dc\u8a0a\u5171\u6709\u516b\u7a2e\uff0c\u5206\u5225\u70ba \u70ba 32\u300164\u3001128\u3001256\u3001512\u30011024\u3002\u5728\u5be6\u9a57\u7d50\u679c\u4e2d\uff0c\u6211\u5011\u5c07\u53ea\u5448\u73fe\u5f97\u5230\u6700\u4f73\u8fa8\uf9fc\uf961\u6642\u7684 N \u503c\u4e4b 77.69 76.08 77.48 76.63 78.21 76.87 75.13 73.84 75.63 77.13 72.91 75.86 77.69 76.08 77.48 76.63 78.21 76.87 75.13 73.84 75.63 77.13 72.91 75.86 75.86 77.73 78.12 80.66 74.32 77.82 AC-LPCC (N=1024) 76.90 77.97 79.42 74.91 75.29 MS-LPCC (N=1024) 76.90 77.97 79.42 74.91 75.29 MS-LPCC (N=1024) 75.86 77.73 78.12 80.66 74.32 77.82 AC-LPCC (N=1024) \u6574\u9ad4\u5be6\u9a57\uf969\u64da\u3002 \u5c0d\u65bc\u7d14\u96dc\u8a0a\u7684\u4f30\u6e2c\u503c [ ] { } p n \uff0c\u6211\u5011\u662f\u4ee5\u5728\u4e2d\u4ecb\u7279\u5fb5\uf96b\uf969\u57df\u4e0a\uff0c\u6bcf\u4e00\u6bb5\u6e2c\u8a66\u8a9e\u97f3\u7684\u524d 5 \u500b\u97f3 1. \u76f8\u5c0d\u65bc\u539f\u59cb\u672a\u8655\uf9e4\u7684\u5012\u983b\u8b5c\uf96b\uf969(\uf969\u64da\uf99c\u65bc\u8868\u4e09)\u800c\u8a00\uff0c\u9019\u4e09\u7a2e\u65b0\u7684\u7279\u5fb5\uf96b\uf969\u88dc\u511f\u6cd5\u90fd\u80fd 77.13 79.81 80.44 81.17 78.91 78.71 MS-PLPCC (N=64) 77.36 76.69 79.40 75.56 77.79 AC-PLPCC (N=64) 79.81 80.44 81.17 78.91 78.71 MS-PLPCC (N=64) 77.13 77.36 76.69 79.40 75.56 77.79 AC-PLPCC (N=64) \u5730\u4e0b\u9435\u3001\u4eba\u8072\u3001\u6c7d\uf902\u3001\u5c55\u89bd\u9928\u3001\u9910\u5ef3\u3001\u8857\u9053\u3001\u6a5f\u5834\u3001\u706b\uf902\u7ad9\u7b49\uff0c\u524d\u56db\u7a2e\u6b78\uf9d0\u70ba Set A\uff0c\u5f8c\u56db\u7a2e\u6b78 \u6846\u7576\u4f5c\u8a72\u6bb5\u8a9e\u97f3\u7684\u7d14\u96dc\u8a0a\u97f3\u6846\u3002 \u5920\u5927\u5e45\u63d0\u6607\u8fa8\uf9fc\u7cbe\u78ba\ufa01\uff0c\u610f\u8b02\u5404\u7a2e\uf967\u540c\u7684\u7279\u5fb5\uf96b\uf969\u90fd\u80fd\u85c9\u7531\u9019\u4e09\u7a2e\u65b9\u6cd5\u800c\u63d0\u5347\u5176\u5f37\u5065\u6027\u3002 \uf9d0\u70ba Set B\u3002 \u8a0a\u96dc\u6bd4(signal-to-noise ratio, SNR)\u5247\u6709\u4e03\u7a2e\uff0c\u5206\u5225\u70ba 20dB, 15dB, 10dB, 5dB, 0dB. -5dB \u8207\u5b8c\u5168\u4e7e \u6de8\uf9fa\u614b\u3002 \u5728 LPCC \u8207 PLPCC \uf978\u7a2e\u7279\u5fb5\uf96b\uf969\u64f7\u53d6\u904e\u7a0b\uf9e8\uff0c\u56e0\u70ba\u5176\u5177\u5099\u8a9e\u97f3\u8207\u96dc\u8a0a\u70ba\u7dda\u6027\u76f8\u52a0\u7684\u4e2d\u4ecb 2. \u5728\u5927\u90e8\u5206\u7684\u60c5\u5f62\u4e0b\uff0c\u9019\u4e09\u7a2e\u65b0\u7684\u7279\u5fb5\uf96b\uf969\u88dc\u511f\u6cd5\u7684\u8868\u73fe\u90fd\u512a\u65bc S-CMVN \u8207 U-CMVN\uff0c\u9019\u547c \u4e03\u3001\u7d50\uf941\u8207\u672a\uf92d\u5c55\u671b \u7279 \u5fb5 \uf96b \uf969 \u6709 \uf978 \u7a2e \uff0c \u5206 \u5225 \u70ba \u5f37 \ufa01 \u983b \u8b5c (magnitude spectrum) \u8207 \u81ea \u76f8 \u95dc \u4fc2 \uf969 (autocorrelation \u61c9\uf9ba\u6211\u5011\u4e4b\u524d\u7684\u63a8\uf941\uff1a\uf9dd\u7528\u78bc\u7c3f\uf92d\u4f30\u6e2c\u7279\u5fb5\uf96b\uf969\u7684\u7d71\u8a08\u503c\u76f8\u8f03\u65bc\uf9dd\u7528\u6574\u6bb5\u6216\u5206\u6bb5\u7684\u65b9\u5f0f\u4f30 \u672c\uf941\u6587\u63d0\u51fa\u4e09\u7a2e\u4ee5\u865b\u64ec\u96d9\u901a\u9053\u78bc\u7c3f\u70ba\u57fa\u790e\u7684\u7279\u5fb5\uf96b\uf969\u88dc\u511f\u6cd5\uff0c\u5206\u5225\u70ba\u5012\u983b\u8b5c\u7d71\u8a08\u88dc\u511f\u6cd5 coefficients)\uff0c\u56e0\u6b64\u5728\u5be6\u9a57\u7d50\u679c\u4e2d\u6211\u5011\u4ee5 MS-\u8207 AC-\u5206\u5225\u4ee3\u8868\u4e4b\u3002 \u6e2c\uf901\uf92d\u7684\u7cbe\u78ba\u3002 (CSC)\u3001\u7dda\u6027\u6700\u5c0f\u5e73\u65b9\u56de\u6b78\u6cd5(LLS)\u8207\u4e8c\u6b21\u6700\u5c0f\u5e73\u65b9\u56de\u6b78\u6cd5(QLS)\uff0c\u500b\u5225\u4f5c\u7528\u65bc\u56db\u7a2e\u8a9e\u97f3\u7279\u5fb5\uf96b</td></tr><tr><td>(\u4e09) \u7279\u5fb5\uf96b\uf969\u7684\u8a2d\u5b9a\u8207\u8fa8\uf9fc\u7cfb\u7d71\u7684\u8a13\uf996 \u672c\uf941\u6587\u5171\u4f7f\u7528\u56db\u7a2e\u7279\u5fb5\uf96b\uf969\u5206\u5225\u70ba\u6885\u723e\u5012\u983b\u8b5c\u4fc2\uf969(MFCC)\u3001\u81ea\u76f8\u95dc\u6885\u723e\u5012\u983b\u8b5c\u4fc2\uf969 3. \u96d6\u7136\u9019\u4e09\u7a2e\u7279\u5fb5\uf96b\uf969\u88dc\u511f\u6cd5\u4f5c\u7528\u65bc\u56db\u7a2e\u7279\u5fb5\uf96b\uf969\u4e0a\u7684\u5be6\u9a57\u7d50\u679c\uff0c\u6240\u5f97\u5230\u6700\u4f73\u8fa8\uf9fc\uf961\u6642\u7684 N \uf969\uff1a\u6885\u723e\u5012\u983b\u8b5c\u4fc2\uf969(MFCC)\u3001\u81ea\u76f8\u95dc\u6885\u723e\u5012\u983b\u8b5c\u4fc2\uf969(AMFCC)\u3001\u7dda\u6027\u9810\u6e2c\u5012\u983b\u8b5c\u4fc2\uf969(LPCC) \u8207 \uf9d1\u3001\u5be6\u9a57\u7d50\u679c\u8207\u5206\u6790 \u9996\u5148\uff0c\u8868\u4e09\u8207\u8868\u56db\u5206\u5225\u70ba\u6574\u6bb5\u5f0f\u5012\u983b\u8b5c\u5e73\u5747\u8207\u8b8a\uf962\uf969\u6b63\u898f\u5316\u6cd5(U-CMVN)\u8207\u5206\u6bb5\u5f0f\u5012\u983b\u8b5c \u611f\u77e5\u7dda\u6027\u9810\u6e2c\u5012\u983b\u8b5c\u4fc2\uf969(PLPCC)\u4e0a\u3002\u6211\u5011\u767c\u73fe\uff0c\u4ee5\u865b\u64ec\u96d9\u901a\u9053\u78bc\u7c3f\u70ba\u57fa\u790e\u4e4b\u7279\u5fb5\uf96b\uf969\u88dc\u511f\u6cd5\uff0c</td></tr></table>", |
| "type_str": "table", |
| "html": null |
| } |
| } |
| } |
| } |
