基于eemd和ica的语音去噪方法-journalofnortheasternuniversity.pdfVIP

第32 卷第11期 东 北 大 学 学 报 ( 自 然 科 学 版 ) V o l. 32, No. 11 2 0 1 1 年 11 月 Journal of Nort heastern U n vers ty( N atural Sc ence) N o v. 2 0 1 1 EEMDICA 李晶皎, 安 冬, 王 骄 ( , 110819) : . , EEM D ICA , ( EEM D) ( I M F) , ( EM D) ; ( ICA) ; , . , - 10dB , 2. 74 1 2 dB . : ; ; ; ; : N 945 . 13 : A : 1005- 3026( 20 11) 11- 1554- 04 Speech Denoising Method Based on the EEMD and ICA Approaches L I Jing-j iao, A N Dong, WAN G Jiao ( School of Informat on Sc ence Eng neer ng, Northeaster n U n v ers ty, Shenyang 110819, Ch na. Correspond ng author : AN Dong , E- ma l : 249350656 @ qq. com) Abstract: Speech deno s ng technology s one of t he key problems n t he pract cal appl cat on of speech recogn t on syst ems. S nce speech s g nals are nonst at onary, speech s gnal conta ned ch rp w as decomposed nto several ntr ns c mode funct ons ( IM F) w th t he met hod of ensemble emp r cal mode decompos t on ( EEM D) . A t the same t me, t el m nat ed the model m x superpos t on phenomenon wh ch usually came out n process ng speech s gnal w th the algor t hm of em p r cal mode decompos t on ( EM D) . Af ter t hat , several ef fect ve speech s gnal component s w ere separated f rom nt r ns c mode f unct on through t he algor thm of mproved ndependent component analys s ( ICA) . F nally , reconstructed t hem n t he purpose of no se el m nat on. T he result show ed that t he new speech deno s ng method proposed above mproves SN R up to 2741 2 dB n t he cond t on that - 10 dB SN R veh cle nter or no se. Key words: emp r c