Voice API test bench

Status

APIchecking

API key–

llama.cpp–

Models–

Settings

API key

Voice Speed

Max reply tokens

System prompt (optional override)

Your voice

Record 5–10 seconds of one person speaking clearly in a quiet room (the best ~5 seconds are kept, to keep replies fast). The assistant then replies in that voice. Only clone voices you have permission to use.

The quick brown fox jumps over the lazy dog. I'm recording this so my assistant can talk back in my own voice.

Name

Or upload a clip (WAV, MP3, M4A…)

Voice chat

Tap the mic, ask something, tap again to send.
Or type below.

Tap to talk

Or send an audio file through the full chain

Test each model on its own

Text → speech (uses the selected voice)

Speech → text (Moonshine)

LLM only, streamed (llama.cpp)

Turn this page off with TEST_UI=false once you're happy.