Prosody Generation Model for TTS Systems