In this thesis I suggest and evaluate an algorithm for the unsupervised segmentation of audio speech streams. Specific attention will be paid to the developmental psychology of human infants, who learn to perform this task at an early age. The goal will be to both suggest an algorithm inspired by the human distributional segmentation mechanism, and to evaluate the performance of that model on acoustic speech. I will focus on the audio domain, in contrast to a great body of previous work devoted to the unsupervised segmentation of text. The algorithm presented is used to reproduce a famous series of infant experiments, and shown to perform similarly to the children. It is also used to segment a large audio corpus, which it does with accuracy significantly better than chance. Finally, improvements to the acoustic model and segmentation algorithm are outlined, implemented and tested, demonstrating the potential for future development of the system.