Input text is first normalised, separated into syllables, detecting its stress, and converted into a sequence of phonemes: http://atrilla.net/data/files/micnlp

Input text is first normalised, separated into syllables, detecting its stress, and converted into a sequence of phonemes: http://atrilla.net/data/files/micnlp09.pdf http://atrilla.net/data/files/micnlp09.pdf Then, a waveform generator creates the audible sounds based on this phoneme transcription (this latter part frequently differentiates the essential TTS synthesis technology: formant-based, concatenative, Hidden Markov Model-based, hybrid approaches, etc).