Input Source Ready
🤖 Gemini Nano Prompt Workbench
On-Device LLM
Temperature
0.30
Top-K
3
Semantic VAD & Sliding Window Tuning
0.50
500 ms
1500 ms
EOU Probability Curve P(Complete) Sub-15ms inference
1. Raw Stream & Semantic VAD
IDLE
Waiting for audio words to stream...
TurnSense EOU Confidence
0.000
2. Semantic VAD Chunk Stream
0 complete sentences detected
No complete sentence chunks detected yet. Speak to see detected sentence chunks appear here line-by-line...
3. Chrome Gemini Nano Smoothed Stream
Sliding Window: 0 turns
TTFT: -- ms
-- tok/s
Smoothed speech text will stream live here via Gemini Nano Prompt API...
Ready to process chunks
4. Gemini Nano Prompt & Sliding Window Inspector
Live Debug Log
Current Active Spoken Chunk
--
Sliding Context Memory (Previous Turns)
Turn 1 (Initial sentence - no prior context needed)
Exact Prompt Payload Sent to Nano
~0 tokens
Waiting for prompt dispatch...
5. Before vs After Comparison
Raw ASR Transcript
--
Gemini Nano Smoothed Output
--
Word-by-Word Telemetry Log
| # | Word Token | Audio Time | TurnSense | P(EOU) | Decision |
|---|---|---|---|---|---|
| No events recorded yet | |||||