Speech Synthesis Using Unsupervised Learning