I am building libraries for predicting which pronunciation of a homograph to use for in a text-to-speech application. E.g., when we go to synthesize "bass", sho

I am building libraries for predicting which pronunciation of a homograph to use for in a text-to-speech application. E.g., when we go to synthesize "bass", should we use the pronunciation that rhymes with "mace" or "pass"?