Research Ofkorean Dialect Identification Based On Prosody

Dialect identification is a very important research issue in the field of speech recognition. Up to now, most speech recognition systems are trained by multi-languages, and the ways of dialect identification is lacked. With the rapid development of science and technology and economy, the communication between different dialects areas is more frequent, so the research and application of dialect identification technique become of the utmost importance. In this dissertation, proposed an approach of the spoken Korean dialect identification, it was using prosodic features and SVM to automatically identify dialect. Dialect identification could concrete application and implementation of language identification in the field of Korean dialect, and it could be a pre-process technology of Korean language identification. Furthermore, the proposed approach has a widely application prospect in fields of automatic machine translation front-end processing, multi-language information service, artificial consulting services and speech engineering, and etc.Firstly, according to characteristic of speech is a kind of quasi-periodic signal, shift delta coefficients were extracted from sequence of pitch frequency features extraction, framing and cepstrum computation.Secondly, SVM was selected as the classifier of Korean dialect identification. Combining DPRK and China official spoken Korean corpus into a whole training samples to train SVM_SY classifier. On other hand, samples constructed with corpus of China and Republic of Korea were used to train SYM_CH classifier.Finally, accuracy rate of dialect identification method was estimated by voting mechanism applied for three consecutive times of discriminant results.Experimental results show that, dialect identification method based on the shift delta coefficient of pitch frequency has achieved 92% of accuracy rate. So, the method proposed in this dissertation could solve effectively issue of China, Korea and DPRK official spoken Korean dialect identification, and was proved its rationality and validity.