Abstract:
The Chinese and English cross-lingual adaptation algorithms are studied, which are based on state mapping method. Two modes of state mapping method applied in cross-lingual adaptation are compared and analyzed. Experiments show that the adaptation algorithms can make sense, and that CSMAPLR is better than CMLLR algorithms. Furthermore, voice quality is better in transform mode than in data mode. And it needs high consistency of speaker characteristics of two original language models. On the contrary, speaker similarity in data mode is better.