Designing a speech prompt is similar to designing any other portion of a speech interface. You must decide whether speech is really the best way to interact with your user. If it is, you must decide whether the prompt should be a TTS utterance or a pre-recorded prompt.
Speech is generally appropriate for relatively short prompts and for tabular information, like weather reports or movie schedules. Long speech prompts confuse the new user and frustrate experts.
Avoid mixing pre-recorded prompts and TTS, especially in the same greeting. Modern TTS engines sound perfectly acceptable on their own, but they do not sound like recordings of any one individual voice. By mixing the two types of prompt, you can create unintended, jarring contrasts that confuse or aggravate your user.