A method of generating a synthetic voice by capturing audio data, cutting it into discrete phoneme and pitch segments, forming superior phoneme and pitch segments by averaging segments having similar phoneme, pitch, and other sound qualities, and training neural networks to correctly concatenate the segments.