Livoice
Talk to GPT-Live in a converted voice.
Voice
Maz
Maz (large model, +150 ms)
Kinu (fine-tuned)
Kinu (zero-shot)
Kinu (large model, +150 ms)
Trump
Japanese girl (azuma)
Anime girl (teio)
No conversion (raw GPT-Live)
Upstream GPT-Live voice (the voice being converted)
auto (recommended for the chosen voice)
marin (female)
coral (female)
sage (female)
shimmer (female)
alloy (neutral)
cedar (male)
ash (male)
echo (male)
verse (male)
ballad (male, British)
Instructions to the model
Start talking
Or type instead of talking (spoken to the model by the Mac's voice)
Send
mute my microphone (send silence)
I'm on headphones (turns echo cancellation off; set before pressing Start)
fast interrupt: cut playback the moment I start talking (only active in headphones mode)
Headphones help a lot: without them the model hears itself through the speaker and interrupts itself.
Expect about a second between when you stop talking and when it answers. That's mostly the model.