@nprofile1q... a tts engine should not require hardware acceleration. it is a bloody waveform generator, but you're just assuming tts engines have to be neural-based. they don't. look at DecTalk, look at IBM TTS / Eloquence, look at something like that