This paper aims to build a Bangla speech sentence recognition system by Hidden Markov Model (HMM). This system includes two phases; such as, a feature extraction phase to generate speech features from the Bangla speech sentence and a recognition phase to identify the Bangla speech sentence. The Mel Frequency Cepstral Coefficients (MFCCs) have been used to generate the features from the input Bangla speech sentences. These MFCCs features were used in the HMM based speech recognizer to identify the Bangla speech sentences. Train data with the different types of HMM training algorithm. In both training and testing process, one hidden Markov model for each sentence has been implemented. The models were trained with labeled training data, and the classification was performed by passing the features to each model and selected the best match. The development and experiments were done on MATLAB 2010 and the learning behavior of the algorithms was tested on different Bangla speech sentences from different speakers.