The Nitech-NAIST HMM-Based Speech Synthesis System for the Blizzard Challenge 2006

The present paper describes an HMM-based speech synthesis system developed by the Nitech-NAIST group for the Blizzard Challenge 2006 (Nitech-NAIST-HTS 2006). To achieve improvements over the 2005 system (Nitech-HTS 2005), new features such as MGC-LSP, MLLT, and full covariance GV pdf are investigated. Subjective listening test results show that combining mel-cepstral coefficients, MLLT and full covariance GV pdf achieved the best score and the performance of the 2006 system is significantly better than that of the 2005 system. Results of the Blizzard Challenge evaluations reveal that Nitech-NAIST-HTS 2006 is still competitive even a relatively large amount of training data is used.