Local Voice LLM

VAD Idle

STT Idle

LLM Idle

TTS Idle

Input

Mic off

Waiting for speech.

Input level

Audio idle

Load the models, start the microphone, and speak naturally.

Device

Checking runtime Adapter probe pending.

Build metadata pending.

LLM STT TTS voice TTS steps 2 VAD silence 480 ms Partial ASR previews

Real mic validation Load models to collect 3 rows Use the mic series and say: "What app is this?"

Hardware WebGPU validation Checking adapter Adapter probe pending.

No benchmark runs yet.

Run	Stack	ASR	STT WER	VAD close	Prompt	1st token	TTS queued	TTS synth	1st audio	End → audio	Audio done	Decode	LLM OK	Transcript	Output
No benchmark runs yet.