Research On Mandarin-Tibetan Cross-lingual Speech Synthesis

In recent years,cross-lingual speech synthesis has become a hot topic of research.China is a country with a large number of minority languages.English research shows that cross-lingual speech synthesis can be realized in English and minority languages.Before this we have realized a hidden Markov model(HMM)-based Mandarin-Tibetan cross-lingual speech synthesis,but there are two problems in this Mandarin-Tibetan cross-lingual speech synthesis.How to realize HMM-based Mandarin-Tibetan cross-lingual emotional speech synthesis to improve the emotional expression of synthesized Mandarin and Tibetan speech? In recent years,deep learning has been successfully applied in speech synthesis.Can deep learning improve the synthesized Mandarin and Tibetan speech quality from Mandarin-Tibetan cross-lingual speech synthesis? For the first problem,the thesis presents a hidden Markov model(HMM)-based Mandarin-Tibetan cross-lingual emotional speech synthesis by using an emotional Mandarin speech corpus.For the second problem,the thesis realizes a deep neural network(DNN)-based Mandarin-Tibetan bilingual speech synthesis.The main works and originalities of the thesis are as follows:Firstly,the thesis presents a HMM-based Mandarin-Tibetan cross-lingual emotional speech synthesis by using an emotional Mandarin speech corpus.According to the similarities in pronunciation and emotional expression between Mandarin and Tibetan,this thesis has realized three kinds of HMM-based Mandarin-Tibetan cross-lingual emotional speech synthesis by using an emotional Mandarin speech corpus.Subjective evaluations and objective tests show that all three kinds can synthesize high speech quality of the synthesized Mandarin and Tibetan emotional speech.Secondly,the thesis proposes a method of DNN-based Mandarin and Tibetan bilingual speech synthesis.It realized a DNN-based Mandarin and Tibetan bilingual speech synthesis by using DNN instead of HMM for acoustic model training.Subjective and objective tests show that synthesized Tibetan speech by the proposed method is not only better than the HMM-based Mandarin-Tibetan cross-lingual speech synthesis but also better than the DNN-based Tibetan speech synthesis that only uses Tibetan training corpus.Therefore the proposed method can be used for minority language speech synthesis that lacks speech resources.