Whole call
One clock for the call: caller VAD turns (amber = noisy mode), caller turns from STT (red = cleared), every STT history edit, LLM, TTS, agent audio and router/tool work. Click or drag to pick the detail window.
VAD turnCaller turnCleared turn / words replacedRewriteLLMTTSAgent audioRouter / tools
VAD, turns and level
—Silero speech probability
shaded = agent audio · bottom tick = VAD in speech stateInput level
dBFS per 32 ms frame · dashed = noise floor in windowSNR vs running noise estimate
dB · energy start needs ≥ 8 dB
ProbabilityEMAStart threshold (adaptive)End thresholdLevelNoise floorSNRAgent audio
Response latency
—VAD speech end → first agent audio per answered turn. VAD end includes the end hangover.
STT / routing waitLLMTTS first audio
STT pass duration
—Each STT pass by type.
Single segmentLimited 1Limited 2Full
VAD distributions
retained framesSpeech probability
—Level by VAD state
—VAD turn lengths
—STT history updates
No history updates yetEvery STT pass that changed a caller turn in the conversation history: new text, rewrites with a word diff, clears (background or echo) and restores.
Conversation
caller turns as STT left them · agent TTS text
{
"status": "waiting for Bedrock calls...",
"message": "Complete Bedrock API calls will appear here in real-time"
}
NOISE
--
BOOKING
idle
Ready
0s
30s
VAD
STT-S
STT-A
GENERAL
PROFILE
REGISTRATION
BASINFO
ALTER
DEPT
EXTRACTION
FALLBACK_LLM
STATE_LLM
BOOKING
TTS
AGENT
SYS
Speaker Priority
(diarized speakers)
Transcriptions
(🔵 VAD | 🔴 Started | 🟢 Completed)
No transcriptions yet
Status:
Processing for STT
1-Segment STT: 0 |
Limited 1 STT: 0 |
Limited 2 STT: 0 | Full
STT:
0
1-Segment:
--ms
△ | Limited 1:
--ms
△ | Limited 2:
--ms
△ | Full:
--ms
△
Audio Files: 0
Audio Tiles
(0)
No audio files captured yet