CONF Lazaridis_SP-7_2014/IDIAP SVR vs MLP for Phone Duration Modelling in HMM-based Speech Synthesis Lazaridis, Alexandros Honnet, Pierre-Edouard Garner, Philip N. HMM-based speech synthesis HSMM explicit duration modelling multilayer perceptron phone duration modelling Support Vector Regression EXTERNAL http://publications.idiap.ch/attachments/papers/2014/Lazaridis_SP-7_2014.pdf PUBLIC http://publications.idiap.ch/index.php/publications/showcite/Lazaridis_Idiap-RR-03-2014 Related documents Speech Prosody 2014 In this paper we investigate external phone duration models (PDMs) for improving the quality of synthetic speech in hidden Markov model (HMM)-based speech synthesis. Support Vector Regression (SVR) and Multilayer Perceptron (MLP) were used for this task. SVR and MLP PDMs were compared with the explicit duration modelling of hidden semi-Markov models (HSMMs). Experiments done on an American English database showed the SVR outperforming the MLP and HSMM duration modelling on objective and subjective evaluation. In the objective test, SVR managed to outperform MLP and HSMM models achieving 15.3% and 25.09% relative improvement in terms of root mean square error (RMSE) respectively. Moreover, in the subjective evaluation test, on synthesized speech, the SVR model was preferred over the MLP and HSMMmodels, achieving a preference score of 35.93% and 56.30%, respectively. REPORT Lazaridis_Idiap-RR-03-2014/IDIAP SVR vs MLP for Phone Duration Modelling in HMM-based Speech Synthesis Lazaridis, Alexandros Honnet, Pierre-Edouard Garner, Philip N. HMM-based speech synthesis HSMM explicit duration modelling multilayer perceptron phone duration modelling Support Vector Regression EXTERNAL http://publications.idiap.ch/attachments/reports/2013/Lazaridis_Idiap-RR-03-2014.pdf PUBLIC Idiap-RR-03-2014 2014 Idiap March 2014 In this paper we investigate external phone duration models (PDMs) for improving the quality of synthetic speech in hidden Markov model (HMM)-based speech synthesis. Support Vector Regression (SVR) and Multilayer Perceptron (MLP) were used for this task. SVR and MLP PDMs were compared with the explicit duration modelling of hidden semi-Markov models (HSMMs). Experiments done on an American English database showed the SVR outperforming the MLP and HSMM duration modelling on objective and subjective evaluation. In the objective test, SVR managed to outperform MLP and HSMM models achieving 15.3% and 25.09% relative improvement in terms of root mean square error (RMSE) respectively. Moreover, in the subjective evaluation test, on synthesized speech, the SVR model was preferred over the MLP and HSMM models, achieving a preference score of 35.93% and 56.30%, respectively.