Design And Implementation Of Malay And Indonesian Cross-lingual Speech Synthesis System

In the globalized environment,multilingual speech synthesis in speech information processing is of great significance.The goal of cross-lingual speech synthesis is to obtain synthesized speech of different speakers and different languages through a system,which has a wide range of application prospects.Indonesian is derived from the Malay language in the northeastern part of Sumatra,and the two languages have a high degree of similarity.In addition,as a second language,English is widely used in countries such as Malaysia and Indonesia.It is very common for English words to appear in Malay and Indonesian sentences.This thesis aims to construct a cross-lingual speech synthesis system for Malay（including English words）and Indonesian（including English words）,and researches on mixed-language text analysis and cross-lingual speech synthesis methods.The main work of the thesis includes:（1）According to the language characteristics of different languages,the respective synthetic units of Malay,Indonesian and English are selected,and their similarities and particularities are analyzed,and the common synthetic units are designed based on the IPA.Unify the synthetic units with the same pronunciation in different languages,and keep the synthetic units with different pronunciations separately.On this basis,the automatic segmentation of the phonemes is completed,and the duration information of each phoneme is obtained.（2）Normalize Malay and Indonesian texts,and use the reverse maximum matching algorithm based on syllable list to complete the syllable division of Malay,Indonesian,and mixed-language texts containing English.The methods of k-means clustering and GMM clustering are used to label the prosodic structure and accent of Malay and Indonesian respectively.（3）According to the phonetic characteristics of Malay,Indonesian,and English,the design of context attributes and problem sets shared by the three languages was completed,and a cross-lingual speech synthesis system based on the HMM model using speaker adaptive algorithm was constructed.Cross-lingual speech synthesis of Malay with English words and Indonesian with English words.（4）To improve the quality of synthesized speech,the cross-lingual speech synthesis baseline system based on the HMM is improved.In view of the limitations of the HMM acoustic model,the DNN model is used to replace the decision tree clustering for improvement,speaker coding is introduced to achieve speaker adaptation,and a cross-lingual speech synthesis system based on DNN is constructed.The experimental results show that the cross-lingual speech synthesis system based on the HMM is feasible,and the cross-lingual speech synthesis system based on the DNN introducing speaker coding effectively improves the quality of cross-lingual speech synthesis.