基于LPC和其残差激励模型的汉语语音变换技术.pdfVIP

28 2 V o l . 28 N o. 2 2007 6 Jun. 2007 Journal of M ilitar y Co mmunicat ions T echnolog y X LPC 1 2 1 孙　卓, 岳振军, 宋　巍 1. 3 , 210007; 2. , 211101 : L PC , , , , , : ; ; ; ; : T N 9 12. 3 : A : CN 32-1289 2007 02-0001-04 Chinese Speech Conversion Technology Based on LPC and Residual Model 1 2 1 , - , S UN Zhuo Y UE Zhen j un S ON G W ei 1. P ostg raduate T ea m 3 CE, P L A U ST , Nanj ing 2 10007, China ; 2. Depart ment of Basic Electr onics S, PL A U ST , Nanj ing 2 1110 1, China Abstract: Speech conversion is a new t echnolo gy to chang e the source speaker s features int o t he targ et speakers features in t he speech . n t his paper , t he Chinese speech conversio n sy st em w as div ided int o t hree part s. n t he f ir st and seco nd part s, different GM M w ere used t o transfo rm t he spectral env elo pe LPC and the impulse residual . n t he t hird part , t he Chinese speechs super-seg ment al f eatures were transf orm ed w ith the pit chs gauss mo del and t he T D- PSOLA . Compared w it h the sim ilar arit hm et ic in ex ist ence, t his ar it hmet ic can reg ulat e t he - . transfo rmed speech pro sody by Chinese super segm ent al f eatures T his arit hmetic can transfor m Chinese speech availably, and produce spont aneo us synthesized speech . Key words: Chinese speech conversion; LP C; residual; Gaussian m ixture model; super- segm ent al f eature voice conversio n vo ice transf ormat ion , so ur ce speaker t arg et [ 1] speaker , LPC Linear Predict ion Coefficient , LPC [ 2] LPC Residual , , , X : 2006-09-13; : 2007-03-26 : 1981- , , . 2 2007 , , 1 LPC , L SF [ 3] , L SF , , , L SF , LSF , , EM , w i , Li , 2i , i 1, , M w i , Li , 2i , ? x i p i x w iN x ; Li , 2i p i?x M w iN x ; Li , 2i i 1 i w i L SF x L SF , y M t ts s s - 1 s y f x [ L + 2i 2i x - L ] ?p i?x i 1 2 LPC , L SF , LPC N p j , i mj M i N i 1 p j , i k 1 M i