This paper describes a new presentation of continuous speech in terms of the probability of all phoneme types as a function of time. The presentation is called a phoneme probability presentation (PPP) and can be used for phoneme segmentation or phoneme lattice production. As a technique to produce the PPP, we have employed hidden Markov models (HMM) with time duration information. This information is essential to produce the PPP and is effective in English phoneme recognition. With this information the HMMs of all the phoneme types can compute their probability in parallel and in time synchronism. The PPP can serve as phoneme filters which can produce phoneme probability from continuous speech.