We describe a new speech production model for improving the quality of synthetic vowel speech. The strong point of this model is that the source model has a continuous harmonic structure in the time domain and the frequency domain. This report provides a comparative study of the ordinary source model, such as impulse and Rosenberg one, and the new model. By listening test, it is confirmed that our model produces high quality synthetic speech which is better as compared with synthetic ones using ordinary models. Thus, we understand that the really continuous harmonic structure in speech spectrum significantly contributes to synthetic speech quality.