| 摘要 |
PROBLEM TO BE SOLVED: To provide a method for acoustic model generation that realizes learning and generation of an acoustic model with a tone-combined type automatically provided.SOLUTION: A method for acoustic model generation in the invention includes a model learning step, a tone-combined type extraction step and a tone-combined type model learning step. In the model learning step, a voice synthesis HMM is learned on the basis of input of learning voice data including a pitch parameter and a spectrum parameter and input of speech information including phoneme segmentation information and accent information of the speech data. In the tone-combined type extraction step, a voice parameter having the phoneme segmentation information identical to the speech information is generated from the voice synthesis HMM, and the tone-combined type between accent phrases is extracted using the voice parameter and the parameter of the learning voice data. In the tone-combined type model learning step, the voice synthesis HMM is learned on the basis of input of the voice data, the speech information and the tone-combined type, and an acoustic model is output. |
