In this paper, we analyze the effects of several factors and configuration choices encountered during training and model construction when we want to obtain better and more stable adaptation in HMM-based speech synthesis. We then propose a new adaptation algorithm called constrained structural maximum a posteriori linear regression (CSMAPLR) whose derivation is based on the knowledge obtained in this analysis and on the results of comparing several conventional adaptation algorithms. Here, we investigate six major aspects of the speaker adaptation: initial models; the amount of the training data for the initial models; the transform functions, estimation criteria, and sensitivity of several linear regression adaptation algorithms; and combination algorithms. Analyzing the effect of the initial model, we compare speaker-dependent models, gender-independent models, and the simultaneous use of the gender-dependent models to single use of the gender-dependent models. Analyzing the effect of the transform functions, we compare the transform function for only mean vectors with that for mean vectors and covariance matrices. Analyzing the effect of the estimation criteria, we compare the ML criterion with a robust estimation criterion called structural MAP. We evaluate the sensitivity of several thresholds for the piecewise linear regression algorithms and take up methods combining MAP adaptation with the linear regression algorithms. We incorporate these adaptation algorithms into our speech synthesis system and present several subjective and objective evaluation results showing the utility and effectiveness of these algorithms in speaker adaptation for HMM-based speech synthesis.

No abstract

No abstract

Spoken digit recognition using autocorrelation coefficients (PAC or k parameters) is investigated with a digital computer. Speech samples are 10 spoken Japanese digits uttered by a single male speaker. Each digit is uttered 10 times. Each sample is analyzed at every 10 msec to give time series of PAC. Principle of recognition is a simple pattern matching with nonlinear time adjustment by DP procedure. When the dimension of PAC is greater than 4, recognition score is fairly good, but does not exceed the value of 98%. The cause of error seems to be the accuracy deterioration of computation of PAC. The computation of PAC is strongly correlated to a signal amplitude. Weighting PAC by amplitude information, the recognition score is improved and reaches at 100%, with the dimension of PAC greater than 4. When PAC is computed by fixed-point arithmetic with short accumulator length, the score become worse with the increase of the dimension. Further improvement of recognition score and time is expected by the variable use of PAC dimension controlled by the computation accuracy.

No abstract

No abstract