It is important how to estimate the coefficients of speech synthesis filter by using a suitable estimation method and determine an adequate excitation input to the synthesis filter to improve the sound quality in speech synthesis. In this paper, we evaluate the sound quality of ARMA speech synthesis with multipulse excitation by signal-to-noise ratio. The ARMA parameters of the synthesis filter are estimated by an adaptive algorithm which smoothes the estimated parameters through modeled time-varying ARMA parameters. The results show that the SNR have been improved in the ARMA synthesis that compared with the AR synthesis by using the same adaptive estimation method. Furthermore, the results suggest the importance of selecting the coefficients of the synthesis filter as the most adequate parameter set in a frame of speech from the estimated parameters by adaptive estimation.