There are two ways that SQLphone can speak a memory. Your flowchart can use either, or both of these methods:
Standard prompts are voice recordings of all the individual numbers, months and days of the week (plus some other words, like "hundred"). By playing standard prompts in correct sequence, any number, date or time can be spoken.
For example: a memory named LightYears might contain the value 10234870. To speak this value to a caller, SQLphone plays appropriate standard prompts in sequence. So the caller hears "ten-million-two-hundred-and-thirty-four-thousand-eight-hundred-and-seventy".
Generally, you will get better results using standard prompts, than using text-to-speech. Why? Because standard prompts use a real human voice (maybe your own voice), not a synthesized voice. However, there are situations where standard prompts cannot be used. For example: when speaking names or addresses.