Let’s talk.
Click the microphone to begin.
Pause when you’re done. Your reply plays aloud.
Click to talkUp to 60 seconds
Type, edit transcript, or upload audio
Up to 10 MB / 90 seconds
Models & Gemini connection
Not connected
The key stays in the local server’s memory until you forget it or restart. Gemini audio sends recordings to Google; local input sends only transcripts when Gemini replies are selected. API usage is billed by your Google plan.
Checking input…Checking replies…
Hokkien voice uses Singapore pronunciation. Gemini audio sends recordings to Google.
Model credits
Hokkien speech uses MERaLiON by A*STAR, built on OmniVoice. Model weights are unmodified; this app adds a fixed synthetic reference speaker. Model licence and notices.