text to speech engines do not work like this. They built the pronunciations of any words on the fly using their phonetic sound library. so 90M is for the sound library I think
If you had 2-300 words, it would be much cheaper to just record the words.
Otherwise, if it's still just dictionary words, it's just a lot easier if the dictionary already breaks down the word into phonemes, and the speech engine then collates the phoneme sounds together. You're not actually using most of the features of a text-to-speech engine, so I don't see why you'd have to pay for the full price tag.
So: how many words? And what's a price that would make you want to buy a license to a simpler engine? :)
