ѧ
JOURNAL OF SOFTWARE
1999ꡡ109ڡVol.10No.91999



ھ߱仯ѧϰеӦ*
͢ۿġCharles X. Ling䡡
ժҪ߱仯ɶԺϳɸȻȵǳҪ.Ϊȡ߼Ƶı仯ɣʵھϵͳSpeechDMѵȡ˽Ϻõѧϰ.Ľ˹ڽеĹ߱仯оԼھSpeechDMϵͳṹҪģ飬ʵ˷.
ؼʡϳ,߱仯,ھ,,ȡ.
ͼšTP311
Data Mining for Mandarin Tone Variation Patterns Learning
ZHU Ting-shao,GAO Wen,GAO Zheng-qi,LI Jin-tao
(Institute of Computing Technology The Chinese Academy of Sciences Beijing 100080)
Charles X. Ling
(Department of Computer Science University of Western Ontario Ontario Canada)
AbstractThe tone variation patterns is very important for speech synthesis. To extract more accurate patterns, a prototype system called SpeechDM has been implemented for learning. In this paper, the relative research and data mining are specified at first, the modules are introduced in detail, and some experimental results are also given.
Key wordsSpeech synthesis, tone variation, data mining, neural network, patterns extraction.
ϳɼǼܽӿڵҪɲ֣Ϊ˺ϳɸȻȡȵƵɹ򣬶ɹܽӦòöķ.ɹҪڵʱֲߵı仯ı仯ʵͣٵȼ棬ߺı仯ȻȵӰΪ1ݣߵı仯ҪƵı仯.
ڼԺģʽܶо.Ϊи15ģʽ2.ïͨͨѧõͨǰڵʱߵĹϵ3.ͨһŮ˵߷,ͳƹɳ14ֵڵģʽ22ִģʽеڵģʽƽģʽ֡ƽģʽ3֡ȥģʽ4֣1.
ĽԿĿǰ߱仯ɴǸѧоóһЩԵʹüϳʱֻṩһЩοںϳɹ׼ȷʹЩ.⣬ʹ˶Щ˹ͳƵõģҲܱ֤ȫ߱仯򣬶ҶЩάҲ.
ݿе֪ʶKDD(knowledge discovery in database)ǽ˹ܺݿ⼼ķչֵһ˵ļǴӴȡŵġӱġЧģܱģʽĸ߼̣4.ھKDDҪĴ׶Σʹߣ5.
˹ȡɹѶԴݽдõɹ򲻹ȫ.ھûѧϰķӴȡõ֪ʶΪˣǽ˽ھӦͨƵ仯ɵоʵһԭϵͳSpeechDM.ִͨʳֵĸߣ5ݣִͨڵĻƵ仯ɵѧϰȡ˽Ϻõʵ.
1SpeechDMϵͳṹ
SpeechDMһݴھϵͳҪԤݹѵھ򼰽ģ.ͼ1SpeechDMϵͳṹ.

ͼ1SpeedDMϵͳṹ
Ϊܹھĸ׶νпƣSpeechDMṩ˽ھĸ̽п.ЩԵѧϰѵĸֲ趨ĳѧϰݼ.
SpeechDMͨݼʵ˶ݵĹ.ݼǶΪĳѧϰݿȡݵ.ϵͳУѵΪ̴߳ڣѵǽм򵥵ĵȣֹͣͣȵ.ݼѵʵַѧϰȡøõѧϰ.
ԤҪǶݽвȡݵĻƵֵйʹʺںѧϰ㷨ʹ.ھģԤݽѵ.ģҪǽԵĽͼλʽʾû.
2Ԥ
ΪѧϰƵ仯ɣȵõĻƵֵ.ԤҪͨݵĴõΪȷĻƵֵʱֵ.
ݿõϳϿCoSS-1ǹ863ֵ֧Ŀ廪ѧѧ뼼ϵйѧԺѧоѧԺоͬɵ.CoSS-11 268еں640ִʣκͳε󣬲ͬ¼ѹκŲ迹.
ʱƵ򷽷ܵõΪ׼ȷĻڣ6ݣ׼ȷĻڶԺѵͲԷǳҪ.CoSS-1еŲ迹ǽ˵񶯲μ¼αȽԸԣԺ׵رÿ.ͨŲĻעõǳ׼ȷĻֵ.
עõĻƵֵı仯ΧܴΪ˱ںб׼.ڲͬڵĻеĳܴЩƵֵҲӦýʱSpeechDMĹ㷨.
룺float *pitch, int pitchlen;/*ĻƵֵͻ*/
float *twpitch, int twlen;/*ĻƵֵ͹*/
0-(twlen-1)еÿֵi,
0-(pitchlen-1)еλloc=pitchlen*i/twlen;
locΪ㣬twleni=pitchloc;
ȡlocΧ4;
òֵlocӦֵvalue,twleni=value.
3ھ
ĻƵֵ仯ڲͬԻ,ͬϼĸµı仯ǲͬ.ǿɽԻӰƵ仯أڵڲͬ·˲ͬı仯Ӷγɶ.˼·ǲ磬ͨѧϰڼڴеӳõƵı仯.
洫(back propagation,BP)һֶṹӳ磬ĿǰӦ㷺һģͣ7.SpeechDMʹ洫лƵ仯ɵѧϰ.Ϊ3㣺㡢㣬֮ȫ.ѧϰȡ0.25SigmoidΪ.
ĿǰSpeechDMҪѧϰִʵĸڵĻƵ仯ɣѧϰִеĸڷڵĵ.Ϊѧϰʹֱѧϰ1͵2ڵĻƵ仯.
ͨǰԤԵõڵʱĻƵֵ͸ִзĻƵֵ.ÿִʣȡÿڵԻĿǰҪʹϢΪԻ.ڵڵʱĻƵֵǰںﹲ4ʹ8ڵʾǰڵǰ4ڵʾ1ڵ4ڵʾ2ڵ磺01001000ʾ1Ϊ2ڵΪһ.
ڵΪִзĻƵֵЩƵֵǾִʽз֡Ƶעʱͱ׼õ.ڵڵĸĳȶͬ.SpeechDM,ȡڵĻƽֵΪĳȣִе1ڵĻ2ڵĻڵƽֵΪڵ.
ݿṩ޷ֱӽѵͲԣȶԭʼݽԤȡڴ˻ϽѵͲ.
Ϊ˵õѵͲݣȶеںִʽзֺͻƵעõƵֵ,ȣڵʱĻƽֵԼִе1ں͵2ڵĻĸƽֵ,ݹȣݽйݿѵͲ.
4ʵ
SpeechDMҪǽʵʷĻƵֵѧϰõĻƵֵбȽϣԽͼηʽʾ.һβԵĽ5.ͼ2УһǴʵĵ1ڵĲԽһʾǵ2ڵĲԽÿһʾͬʱ˹ڵĻƵߡڴзʱĻƵԼԽĻƵ.

Ű(gu3ban3)

(qi3chi3)

β(wei3hua1)

(ke3lian2)

(huo3hua4)
ͼ2ڡзԽĻƵֵ
ͼ25Թ10ڵĲԽʵԿԽʵֵǳӽ.ʹھԵõ߱仯ɵĶԵҪĻƵУֱϳ.ڳʹõĺϳϵͳУ߹еĹб任ܵõҪĻƵ.ѧϰռÿռСʹʱԺܿõҪĻƵУӶӿϳɵٶ.ھеĹȡѧϰ㷨ɵģʹͨԴģݵѧϰԵõΪȫ߱仯Ϊ,ͨѵȽϿԵõõѧϰ.
SpeechDMݴƵһھϵͳʵݿ֪ʶֵҪ裬ȡ˽Ϻõѧϰ.һĹ仯ɵѧϰִеĻƵ仯ɵѧϰ.ϣͨھ򣬵õΪȫĺɹɣߺϳȻȺ.
*оõȻѧйѧԺ˹Ŀ.
ͨѶϵˣ͢ۿ100080,йѧԺ㼼о
߼飺͢ۿ1971ʿҪоΪݿ⼰ӦãھInternetϳ.
ģ1956ʿڣʿʦҪоΪ˹ܣģʽʶͼý弼.
Charles X. Ling1956ʿڣʿʦҪоΪѧϰھӦ.
䣬1971ʦҪоΪӦãھϳ.
Σ1962ʿоԱҪоΪý弼ʵ.
ߵλ͢ۿģ䣬ΣйѧԺ㼼о  100080
Charles X. LingWestern OntarioѧѧϵOntarioô
οף
1ݳ.ȸȻȺתϵͳо۲ʿѧλģ.йѧԺѧо,1995
(Chu Min. Research on Chinese TTS system with high intelligibility and naturalness Ph.D. Thesis. Institute of Acoustics, The Chinese Academy of Sciences, 1995)
2ڼ.ͨе仯.й,1982,6:439449
(Wu Zong-ji. The tone variation in mandarin. Chinese Grammar, 1982,6:439449)
3ï,վ,.ĳʵ.,1984,1
(Lin Mao-can, Yan Jing-zhu, Sun guo-hua. Experiment of the normal accent in Beijing dialect. Dialect, 1984,1)
4Usama M Fayyad, Gregory Piatetsky-Shapiro, Padhraic Smyth et al. Adavance in Knowledge Discovery and Data Mining. Cambridge, MA: AAAI/MIT Press, 1996
5John George H. Enhancements to the data mining process Ph.D. Thesis. Stanford University, 1997
6о,ٻ.źִ.ӹҵ,1990
(Yang Xing-jun, Chi Hui-sheng et al. Speech Signal Digital Process. Beijing: Publishing House of Electronics Industry, 1990)
7ΰ.˹ԭӦ.պѧ,1995
(Wang Wei. Principle of Artificial Neural NetworkRudiment and Implement. Beijing: Beijing University of Aeronautics and Astronautics Press, 1995)
ոڣ1998-05-15޸ڣ1998-10-05
