Analysis of speech, robust feature representations, building voices using noisy data, and adaptation of TTS systems using spoken user feedback