Pitch modeling for Chinese Speech mixed with English word spelling
Wen-Hsing Lai · 2008
A pitch modeling approach for Mandarin Chinese Speech mixed with English word spelling is proposed. Orthogonal transformation was taken to represent pitch period contour. The first coefficient of transformation represents the pitch mean and the others represent the shape. The pitch mean coefficient, which somehow represents the intonation, is modeled. It takes several major factors as additive factors and estimates model parameters by an EM (expectation-maximization) algorithm. Experimental results showed that RMSEs of 0.41 and 0.83 ms in the pitch mean were obtained for the closed and open tests, respectively. The prosodic-state labels produced by the EM algorithm were linguistically meaningful, even with the disturbance of English words.