The document discusses unit selection text-to-speech synthesis in MaryTTS. It provides an overview of unit selection TTS and the MaryTTS implementation. The MaryTTS implementation is modularized with unit selection packages divided into groups like base classes, data, selection/cost functions, and weighting functions. It describes the unit selection process in MaryTTS step-by-step, including loading a voice's properties, selecting units with a Viterbi algorithm, and concatenating units. Finally, it asks how the system could be improved, such as restructuring the codebase.