How to send the AI's output to Mac Say using only llama.cpp?

[classicjazz]

I would like to have an interactive chat session using llama.cpp on a Mac, where only the output from AI is piped to the Mac "say" command, which then uses the voice selected in Accessibility settings (e.g. English Ireland > Voice 2). (I want to type my input)

@ggerganov has an example of this, used in conjunction with whisper.cpp, here:
https://twitter.com/ggerganov/status/1640035474205995011

See also
ggml-org/whisper.cpp@master...talk.llama-coreml#diff-919c16cc39dd4e7769d99b1a2866078795836b3a94285a574f865d13199de72e

Can anyone suggest the best way to output the AI's to invoke "say" in a bash script from main, for the AI output only, without needing whisper.cpp?

Thanks