The speech confabulation model is built based on cogent confabulation to process streaming input data to recover a sentence from speech wav. The speech wav is handled using GMM model to give likelihood of each phone as input of confabulation. To improve the performance, sentence level confabulation is introduced.