Speech Synthesis in Mexican Spanish Using LSP as Voice Parameterization
Carlos Franco, Abel Herrera, Boris Escalante
A voice parameterization using Linear Spectral was implemented to a Mexican Spanish HMM-based Speech Synthesis System. Five phrases were synthesized and statistically validated by applying a MOS test to 30 listeners who analyzed the original voices compared to a synthetic voice. Results were compared to previous work where MFCC was used as voice parameterization, the comparison shows that LSP parameterization is above the mean score and pointed better than MFCC.