I looked at SelectVoice() and GetInstalledVoices() as well as the SpeechSynthesizer class.
I saw a similar question, where one answer says
[...] likely to be related to the SAPI version [...]
and the other one says
[...] definitely not related to SAPI version [...]
If I write a speech synthesizer application, what SAPI version should I require my customers to have? I want it as low as possible, of course, but I don't want to get in trouble with bug reports that affect SAPI versions the managed API does not support.