HMM-based speech synthesis with various speaking styles using model interpolation and adaptation

摘要

This paper presents an approach to realizing various emotional expressions and speaking styles in synthetic speech using HMM-based speech synthesis based on model interpolation and model adaptation techniques. First, we investigate a technique for synthesizing speech with a speaking style in between two different speaking styles by applying a model interpolation technique. From a result of a subjective experiment, it is shown that speech generated from an interpolated model has a speaking style in between two speaking styles. We also investigate a speaking style adaptation technique based on maximum likelihood linear regression. Results of subjective experiments show that it is possible to synthesize speech with a target speaking style by adapting "reading" style model using 50 adaptation sentences.