The linear prediction (LP) HMM does not make the independent and identical distribution (IID) assumption in the traditional HMM; however it often produces unsatisfactory results. In our previous paper [7], both HMMs’ modeling strengths and weaknesses were analyzed and a new combined model of statics-dynamics of speech was proposed. It works with LPHMM as the dynamic part and with the traditional IID-based HMM as the static part; in addition, easy implementation and low cost are preserved. In this paper, an optimal combination using maximum mutual information (MMI) is introduced. Our experiments on speaker-independent continuous speech recognition demonstrated that the combined model achieved better performance than both models.