I haven't tried building festival, to answer the previous query; I'd try espeak instead -- which I think works with Haiku. However, imho, I think festival and espeak have sonically ugly voices.
If text-to-speech support were to come to Haiku, I'd quietly suggest that instead of using espeak or festival to read text dynamically, Haiku could have a compressed set of pre-recorded strings (i.e. "Do you want to shut down the computer now?"), saved as Wave, Ogg Vorbis, etc. sound bytes. These could be from a better open computer-generated TTS engine (i.e. libttspico), or human voice recordings we could do or extract from Voxforge. From there, they could simply be applied to dialogs, etc. as sound events. It wouldn't be as elegant, proper, or what not... but it would work. Plus, it would sound a lot better than Orca and Gespeaker and more like how macOS and Windows 10 presently sound with VoiceOver and Narrator, respectively. Of course, using sound bytes wouldn't be speech synthesis like using espeak would... and it would obviously be very limited in scope. If nothing else, it's just a suggestion to the team/community... but one I thought I'd add into the discussion anyway, for whatever little my opinion is worth.